2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 * Carnegie Mellon requests users of this software to return to
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
64 * Virtual memory mapping module.
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
70 #include <sys/param.h>
71 #include <sys/systm.h>
73 #include <sys/kernel.h>
76 #include <sys/mutex.h>
78 #include <sys/vmmeter.h>
80 #include <sys/vnode.h>
81 #include <sys/racct.h>
82 #include <sys/resourcevar.h>
83 #include <sys/rwlock.h>
85 #include <sys/sysctl.h>
86 #include <sys/sysent.h>
90 #include <vm/vm_param.h>
92 #include <vm/vm_map.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_pageout.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_pager.h>
97 #include <vm/vm_kern.h>
98 #include <vm/vm_extern.h>
99 #include <vm/vnode_pager.h>
100 #include <vm/swap_pager.h>
104 * Virtual memory maps provide for the mapping, protection,
105 * and sharing of virtual memory objects. In addition,
106 * this module provides for an efficient virtual copy of
107 * memory from one map to another.
109 * Synchronization is required prior to most operations.
111 * Maps consist of an ordered doubly-linked list of simple
112 * entries; a self-adjusting binary search tree of these
113 * entries is used to speed up lookups.
115 * Since portions of maps are specified by start/end addresses,
116 * which may not align with existing map entries, all
117 * routines merely "clip" entries to these start/end values.
118 * [That is, an entry is split into two, bordering at a
119 * start or end value.] Note that these clippings may not
120 * always be necessary (as the two resulting entries are then
121 * not changed); however, the clipping is done for convenience.
123 * As mentioned above, virtual copy operations are performed
124 * by copying VM object references from one map to
125 * another, and then marking both regions as copy-on-write.
128 static struct mtx map_sleep_mtx;
129 static uma_zone_t mapentzone;
130 static uma_zone_t kmapentzone;
131 static uma_zone_t vmspace_zone;
132 static int vmspace_zinit(void *mem, int size, int flags);
133 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
135 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
136 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
137 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
138 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
139 vm_map_entry_t gap_entry);
140 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
141 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
143 static void vmspace_zdtor(void *mem, int size, void *arg);
145 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
146 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
148 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
149 vm_offset_t failed_addr);
151 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
152 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
153 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
156 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
159 #define PROC_VMSPACE_LOCK(p) do { } while (0)
160 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
163 * VM_MAP_RANGE_CHECK: [ internal use only ]
165 * Asserts that the starting and ending region
166 * addresses fall within the valid range of the map.
168 #define VM_MAP_RANGE_CHECK(map, start, end) \
170 if (start < vm_map_min(map)) \
171 start = vm_map_min(map); \
172 if (end > vm_map_max(map)) \
173 end = vm_map_max(map); \
181 * Initialize the vm_map module. Must be called before
182 * any other vm_map routines.
184 * Map and entry structures are allocated from the general
185 * purpose memory pool with some exceptions:
187 * - The kernel map and kmem submap are allocated statically.
188 * - Kernel map entries are allocated out of a static pool.
190 * These restrictions are necessary since malloc() uses the
191 * maps and requires map entries.
197 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
198 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
199 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
200 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
201 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
203 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
209 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
213 vmspace_zinit(void *mem, int size, int flags)
218 vm = (struct vmspace *)mem;
221 memset(map, 0, sizeof(*map));
222 mtx_init(&map->system_mtx, "vm map (system)", NULL,
223 MTX_DEF | MTX_DUPOK);
224 sx_init(&map->lock, "vm map (user)");
225 PMAP_LOCK_INIT(vmspace_pmap(vm));
231 vmspace_zdtor(void *mem, int size, void *arg)
235 vm = (struct vmspace *)mem;
236 KASSERT(vm->vm_map.nentries == 0,
237 ("vmspace %p nentries == %d on free", vm, vm->vm_map.nentries));
238 KASSERT(vm->vm_map.size == 0,
239 ("vmspace %p size == %ju on free", vm, (uintmax_t)vm->vm_map.size));
241 #endif /* INVARIANTS */
244 * Allocate a vmspace structure, including a vm_map and pmap,
245 * and initialize those structures. The refcnt is set to 1.
248 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
252 vm = uma_zalloc(vmspace_zone, M_WAITOK);
253 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
254 if (!pinit(vmspace_pmap(vm))) {
255 uma_zfree(vmspace_zone, vm);
258 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
259 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
274 vmspace_container_reset(struct proc *p)
278 racct_set(p, RACCT_DATA, 0);
279 racct_set(p, RACCT_STACK, 0);
280 racct_set(p, RACCT_RSS, 0);
281 racct_set(p, RACCT_MEMLOCK, 0);
282 racct_set(p, RACCT_VMEM, 0);
288 vmspace_dofree(struct vmspace *vm)
291 CTR1(KTR_VM, "vmspace_free: %p", vm);
294 * Make sure any SysV shm is freed, it might not have been in
300 * Lock the map, to wait out all other references to it.
301 * Delete all of the mappings and pages they hold, then call
302 * the pmap module to reclaim anything left.
304 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
305 vm_map_max(&vm->vm_map));
307 pmap_release(vmspace_pmap(vm));
308 vm->vm_map.pmap = NULL;
309 uma_zfree(vmspace_zone, vm);
313 vmspace_free(struct vmspace *vm)
316 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
317 "vmspace_free() called");
319 if (vm->vm_refcnt == 0)
320 panic("vmspace_free: attempt to free already freed vmspace");
322 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
327 vmspace_exitfree(struct proc *p)
331 PROC_VMSPACE_LOCK(p);
334 PROC_VMSPACE_UNLOCK(p);
335 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
340 vmspace_exit(struct thread *td)
347 * Release user portion of address space.
348 * This releases references to vnodes,
349 * which could cause I/O if the file has been unlinked.
350 * Need to do this early enough that we can still sleep.
352 * The last exiting process to reach this point releases as
353 * much of the environment as it can. vmspace_dofree() is the
354 * slower fallback in case another process had a temporary
355 * reference to the vmspace.
360 atomic_add_int(&vmspace0.vm_refcnt, 1);
361 refcnt = vm->vm_refcnt;
363 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
364 /* Switch now since other proc might free vmspace */
365 PROC_VMSPACE_LOCK(p);
366 p->p_vmspace = &vmspace0;
367 PROC_VMSPACE_UNLOCK(p);
370 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt - 1));
372 if (p->p_vmspace != vm) {
373 /* vmspace not yet freed, switch back */
374 PROC_VMSPACE_LOCK(p);
376 PROC_VMSPACE_UNLOCK(p);
379 pmap_remove_pages(vmspace_pmap(vm));
380 /* Switch now since this proc will free vmspace */
381 PROC_VMSPACE_LOCK(p);
382 p->p_vmspace = &vmspace0;
383 PROC_VMSPACE_UNLOCK(p);
389 vmspace_container_reset(p);
393 /* Acquire reference to vmspace owned by another process. */
396 vmspace_acquire_ref(struct proc *p)
401 PROC_VMSPACE_LOCK(p);
404 PROC_VMSPACE_UNLOCK(p);
407 refcnt = vm->vm_refcnt;
409 if (refcnt <= 0) { /* Avoid 0->1 transition */
410 PROC_VMSPACE_UNLOCK(p);
413 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt + 1));
414 if (vm != p->p_vmspace) {
415 PROC_VMSPACE_UNLOCK(p);
419 PROC_VMSPACE_UNLOCK(p);
424 * Switch between vmspaces in an AIO kernel process.
426 * The new vmspace is either the vmspace of a user process obtained
427 * from an active AIO request or the initial vmspace of the AIO kernel
428 * process (when it is idling). Because user processes will block to
429 * drain any active AIO requests before proceeding in exit() or
430 * execve(), the reference count for vmspaces from AIO requests can
431 * never be 0. Similarly, AIO kernel processes hold an extra
432 * reference on their initial vmspace for the life of the process. As
433 * a result, the 'newvm' vmspace always has a non-zero reference
434 * count. This permits an additional reference on 'newvm' to be
435 * acquired via a simple atomic increment rather than the loop in
436 * vmspace_acquire_ref() above.
439 vmspace_switch_aio(struct vmspace *newvm)
441 struct vmspace *oldvm;
443 /* XXX: Need some way to assert that this is an aio daemon. */
445 KASSERT(newvm->vm_refcnt > 0,
446 ("vmspace_switch_aio: newvm unreferenced"));
448 oldvm = curproc->p_vmspace;
453 * Point to the new address space and refer to it.
455 curproc->p_vmspace = newvm;
456 atomic_add_int(&newvm->vm_refcnt, 1);
458 /* Activate the new mapping. */
459 pmap_activate(curthread);
465 _vm_map_lock(vm_map_t map, const char *file, int line)
469 mtx_lock_flags_(&map->system_mtx, 0, file, line);
471 sx_xlock_(&map->lock, file, line);
476 vm_map_entry_set_vnode_text(vm_map_entry_t entry, bool add)
482 if ((entry->eflags & MAP_ENTRY_VN_EXEC) == 0)
484 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
485 ("Submap with execs"));
486 object = entry->object.vm_object;
487 KASSERT(object != NULL, ("No object for text, entry %p", entry));
488 if ((object->flags & OBJ_ANON) != 0)
489 object = object->handle;
491 KASSERT(object->backing_object == NULL,
492 ("non-anon object %p shadows", object));
493 KASSERT(object != NULL, ("No content object for text, entry %p obj %p",
494 entry, entry->object.vm_object));
497 * Mostly, we do not lock the backing object. It is
498 * referenced by the entry we are processing, so it cannot go
503 if (object->type == OBJT_DEAD) {
505 * For OBJT_DEAD objects, v_writecount was handled in
506 * vnode_pager_dealloc().
508 } else if (object->type == OBJT_VNODE) {
510 } else if (object->type == OBJT_SWAP) {
511 KASSERT((object->flags & OBJ_TMPFS_NODE) != 0,
512 ("vm_map_entry_set_vnode_text: swap and !TMPFS "
513 "entry %p, object %p, add %d", entry, object, add));
515 * Tmpfs VREG node, which was reclaimed, has
516 * OBJ_TMPFS_NODE flag set, but not OBJ_TMPFS. In
517 * this case there is no v_writecount to adjust.
519 VM_OBJECT_RLOCK(object);
520 if ((object->flags & OBJ_TMPFS) != 0) {
521 vp = object->un_pager.swp.swp_tmpfs;
527 VM_OBJECT_RUNLOCK(object);
530 ("vm_map_entry_set_vnode_text: wrong object type, "
531 "entry %p, object %p, add %d", entry, object, add));
535 VOP_SET_TEXT_CHECKED(vp);
537 vn_lock(vp, LK_SHARED | LK_RETRY);
538 VOP_UNSET_TEXT_CHECKED(vp);
547 * Use a different name for this vm_map_entry field when it's use
548 * is not consistent with its use as part of an ordered search tree.
550 #define defer_next right
553 vm_map_process_deferred(void)
556 vm_map_entry_t entry, next;
560 entry = td->td_map_def_user;
561 td->td_map_def_user = NULL;
562 while (entry != NULL) {
563 next = entry->defer_next;
564 MPASS((entry->eflags & (MAP_ENTRY_WRITECNT |
565 MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_WRITECNT |
567 if ((entry->eflags & MAP_ENTRY_WRITECNT) != 0) {
569 * Decrement the object's writemappings and
570 * possibly the vnode's v_writecount.
572 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
573 ("Submap with writecount"));
574 object = entry->object.vm_object;
575 KASSERT(object != NULL, ("No object for writecount"));
576 vm_pager_release_writecount(object, entry->start,
579 vm_map_entry_set_vnode_text(entry, false);
580 vm_map_entry_deallocate(entry, FALSE);
587 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
591 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
593 sx_assert_(&map->lock, SA_XLOCKED, file, line);
596 #define VM_MAP_ASSERT_LOCKED(map) \
597 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
599 enum { VMMAP_CHECK_NONE, VMMAP_CHECK_UNLOCK, VMMAP_CHECK_ALL };
601 static int enable_vmmap_check = VMMAP_CHECK_UNLOCK;
603 static int enable_vmmap_check = VMMAP_CHECK_NONE;
605 SYSCTL_INT(_debug, OID_AUTO, vmmap_check, CTLFLAG_RWTUN,
606 &enable_vmmap_check, 0, "Enable vm map consistency checking");
608 static void _vm_map_assert_consistent(vm_map_t map, int check);
610 #define VM_MAP_ASSERT_CONSISTENT(map) \
611 _vm_map_assert_consistent(map, VMMAP_CHECK_ALL)
613 #define VM_MAP_UNLOCK_CONSISTENT(map) do { \
614 if (map->nupdates > map->nentries) { \
615 _vm_map_assert_consistent(map, VMMAP_CHECK_UNLOCK); \
620 #define VM_MAP_UNLOCK_CONSISTENT(map)
623 #define VM_MAP_ASSERT_LOCKED(map)
624 #define VM_MAP_ASSERT_CONSISTENT(map)
625 #define VM_MAP_UNLOCK_CONSISTENT(map)
626 #endif /* INVARIANTS */
629 _vm_map_unlock(vm_map_t map, const char *file, int line)
632 VM_MAP_UNLOCK_CONSISTENT(map);
634 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
636 sx_xunlock_(&map->lock, file, line);
637 vm_map_process_deferred();
642 _vm_map_lock_read(vm_map_t map, const char *file, int line)
646 mtx_lock_flags_(&map->system_mtx, 0, file, line);
648 sx_slock_(&map->lock, file, line);
652 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
656 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
658 sx_sunlock_(&map->lock, file, line);
659 vm_map_process_deferred();
664 _vm_map_trylock(vm_map_t map, const char *file, int line)
668 error = map->system_map ?
669 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
670 !sx_try_xlock_(&map->lock, file, line);
677 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
681 error = map->system_map ?
682 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
683 !sx_try_slock_(&map->lock, file, line);
688 * _vm_map_lock_upgrade: [ internal use only ]
690 * Tries to upgrade a read (shared) lock on the specified map to a write
691 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a
692 * non-zero value if the upgrade fails. If the upgrade fails, the map is
693 * returned without a read or write lock held.
695 * Requires that the map be read locked.
698 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
700 unsigned int last_timestamp;
702 if (map->system_map) {
703 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
705 if (!sx_try_upgrade_(&map->lock, file, line)) {
706 last_timestamp = map->timestamp;
707 sx_sunlock_(&map->lock, file, line);
708 vm_map_process_deferred();
710 * If the map's timestamp does not change while the
711 * map is unlocked, then the upgrade succeeds.
713 sx_xlock_(&map->lock, file, line);
714 if (last_timestamp != map->timestamp) {
715 sx_xunlock_(&map->lock, file, line);
725 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
728 if (map->system_map) {
729 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
731 VM_MAP_UNLOCK_CONSISTENT(map);
732 sx_downgrade_(&map->lock, file, line);
739 * Returns a non-zero value if the caller holds a write (exclusive) lock
740 * on the specified map and the value "0" otherwise.
743 vm_map_locked(vm_map_t map)
747 return (mtx_owned(&map->system_mtx));
749 return (sx_xlocked(&map->lock));
753 * _vm_map_unlock_and_wait:
755 * Atomically releases the lock on the specified map and puts the calling
756 * thread to sleep. The calling thread will remain asleep until either
757 * vm_map_wakeup() is performed on the map or the specified timeout is
760 * WARNING! This function does not perform deferred deallocations of
761 * objects and map entries. Therefore, the calling thread is expected to
762 * reacquire the map lock after reawakening and later perform an ordinary
763 * unlock operation, such as vm_map_unlock(), before completing its
764 * operation on the map.
767 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
770 VM_MAP_UNLOCK_CONSISTENT(map);
771 mtx_lock(&map_sleep_mtx);
773 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
775 sx_xunlock_(&map->lock, file, line);
776 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
783 * Awaken any threads that have slept on the map using
784 * vm_map_unlock_and_wait().
787 vm_map_wakeup(vm_map_t map)
791 * Acquire and release map_sleep_mtx to prevent a wakeup()
792 * from being performed (and lost) between the map unlock
793 * and the msleep() in _vm_map_unlock_and_wait().
795 mtx_lock(&map_sleep_mtx);
796 mtx_unlock(&map_sleep_mtx);
801 vm_map_busy(vm_map_t map)
804 VM_MAP_ASSERT_LOCKED(map);
809 vm_map_unbusy(vm_map_t map)
812 VM_MAP_ASSERT_LOCKED(map);
813 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
814 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
815 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
821 vm_map_wait_busy(vm_map_t map)
824 VM_MAP_ASSERT_LOCKED(map);
826 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
828 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
830 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
836 vmspace_resident_count(struct vmspace *vmspace)
838 return pmap_resident_count(vmspace_pmap(vmspace));
842 * Initialize an existing vm_map structure
843 * such as that in the vmspace structure.
846 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
849 map->header.eflags = MAP_ENTRY_HEADER;
850 map->needs_wakeup = FALSE;
853 map->header.end = min;
854 map->header.start = max;
856 map->header.left = map->header.right = &map->header;
867 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
870 _vm_map_init(map, pmap, min, max);
871 mtx_init(&map->system_mtx, "vm map (system)", NULL,
872 MTX_DEF | MTX_DUPOK);
873 sx_init(&map->lock, "vm map (user)");
877 * vm_map_entry_dispose: [ internal use only ]
879 * Inverse of vm_map_entry_create.
882 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
884 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
888 * vm_map_entry_create: [ internal use only ]
890 * Allocates a VM map entry for insertion.
891 * No entry fields are filled in.
893 static vm_map_entry_t
894 vm_map_entry_create(vm_map_t map)
896 vm_map_entry_t new_entry;
899 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
901 new_entry = uma_zalloc(mapentzone, M_WAITOK);
902 if (new_entry == NULL)
903 panic("vm_map_entry_create: kernel resources exhausted");
908 * vm_map_entry_set_behavior:
910 * Set the expected access behavior, either normal, random, or
914 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
916 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
917 (behavior & MAP_ENTRY_BEHAV_MASK);
921 * vm_map_entry_max_free_{left,right}:
923 * Compute the size of the largest free gap between two entries,
924 * one the root of a tree and the other the ancestor of that root
925 * that is the least or greatest ancestor found on the search path.
927 static inline vm_size_t
928 vm_map_entry_max_free_left(vm_map_entry_t root, vm_map_entry_t left_ancestor)
931 return (root->left != left_ancestor ?
932 root->left->max_free : root->start - left_ancestor->end);
935 static inline vm_size_t
936 vm_map_entry_max_free_right(vm_map_entry_t root, vm_map_entry_t right_ancestor)
939 return (root->right != right_ancestor ?
940 root->right->max_free : right_ancestor->start - root->end);
944 * vm_map_entry_{pred,succ}:
946 * Find the {predecessor, successor} of the entry by taking one step
947 * in the appropriate direction and backtracking as much as necessary.
948 * vm_map_entry_succ is defined in vm_map.h.
950 static inline vm_map_entry_t
951 vm_map_entry_pred(vm_map_entry_t entry)
953 vm_map_entry_t prior;
956 if (prior->right->start < entry->start) {
958 prior = prior->right;
959 while (prior->right != entry);
964 static inline vm_size_t
965 vm_size_max(vm_size_t a, vm_size_t b)
968 return (a > b ? a : b);
971 #define SPLAY_LEFT_STEP(root, y, llist, rlist, test) do { \
973 vm_size_t max_free; \
976 * Infer root->right->max_free == root->max_free when \
977 * y->max_free < root->max_free || root->max_free == 0. \
978 * Otherwise, look right to find it. \
981 max_free = root->max_free; \
982 KASSERT(max_free == vm_size_max( \
983 vm_map_entry_max_free_left(root, llist), \
984 vm_map_entry_max_free_right(root, rlist)), \
985 ("%s: max_free invariant fails", __func__)); \
986 if (max_free - 1 < vm_map_entry_max_free_left(root, llist)) \
987 max_free = vm_map_entry_max_free_right(root, rlist); \
988 if (y != llist && (test)) { \
989 /* Rotate right and make y root. */ \
994 if (max_free < y->max_free) \
995 root->max_free = max_free = \
996 vm_size_max(max_free, z->max_free); \
997 } else if (max_free < y->max_free) \
998 root->max_free = max_free = \
999 vm_size_max(max_free, root->start - y->end);\
1003 /* Copy right->max_free. Put root on rlist. */ \
1004 root->max_free = max_free; \
1005 KASSERT(max_free == vm_map_entry_max_free_right(root, rlist), \
1006 ("%s: max_free not copied from right", __func__)); \
1007 root->left = rlist; \
1009 root = y != llist ? y : NULL; \
1012 #define SPLAY_RIGHT_STEP(root, y, llist, rlist, test) do { \
1014 vm_size_t max_free; \
1017 * Infer root->left->max_free == root->max_free when \
1018 * y->max_free < root->max_free || root->max_free == 0. \
1019 * Otherwise, look left to find it. \
1022 max_free = root->max_free; \
1023 KASSERT(max_free == vm_size_max( \
1024 vm_map_entry_max_free_left(root, llist), \
1025 vm_map_entry_max_free_right(root, rlist)), \
1026 ("%s: max_free invariant fails", __func__)); \
1027 if (max_free - 1 < vm_map_entry_max_free_right(root, rlist)) \
1028 max_free = vm_map_entry_max_free_left(root, llist); \
1029 if (y != rlist && (test)) { \
1030 /* Rotate left and make y root. */ \
1035 if (max_free < y->max_free) \
1036 root->max_free = max_free = \
1037 vm_size_max(max_free, z->max_free); \
1038 } else if (max_free < y->max_free) \
1039 root->max_free = max_free = \
1040 vm_size_max(max_free, y->start - root->end);\
1044 /* Copy left->max_free. Put root on llist. */ \
1045 root->max_free = max_free; \
1046 KASSERT(max_free == vm_map_entry_max_free_left(root, llist), \
1047 ("%s: max_free not copied from left", __func__)); \
1048 root->right = llist; \
1050 root = y != rlist ? y : NULL; \
1054 * Walk down the tree until we find addr or a gap where addr would go, breaking
1055 * off left and right subtrees of nodes less than, or greater than addr. Treat
1056 * subtrees with root->max_free < length as empty trees. llist and rlist are
1057 * the two sides in reverse order (bottom-up), with llist linked by the right
1058 * pointer and rlist linked by the left pointer in the vm_map_entry, and both
1059 * lists terminated by &map->header. This function, and the subsequent call to
1060 * vm_map_splay_merge_{left,right,pred,succ}, rely on the start and end address
1061 * values in &map->header.
1063 static __always_inline vm_map_entry_t
1064 vm_map_splay_split(vm_map_t map, vm_offset_t addr, vm_size_t length,
1065 vm_map_entry_t *llist, vm_map_entry_t *rlist)
1067 vm_map_entry_t left, right, root, y;
1069 left = right = &map->header;
1071 while (root != NULL && root->max_free >= length) {
1072 KASSERT(left->end <= root->start &&
1073 root->end <= right->start,
1074 ("%s: root not within tree bounds", __func__));
1075 if (addr < root->start) {
1076 SPLAY_LEFT_STEP(root, y, left, right,
1077 y->max_free >= length && addr < y->start);
1078 } else if (addr >= root->end) {
1079 SPLAY_RIGHT_STEP(root, y, left, right,
1080 y->max_free >= length && addr >= y->end);
1089 static __always_inline void
1090 vm_map_splay_findnext(vm_map_entry_t root, vm_map_entry_t *rlist)
1092 vm_map_entry_t hi, right, y;
1095 hi = root->right == right ? NULL : root->right;
1099 SPLAY_LEFT_STEP(hi, y, root, right, true);
1104 static __always_inline void
1105 vm_map_splay_findprev(vm_map_entry_t root, vm_map_entry_t *llist)
1107 vm_map_entry_t left, lo, y;
1110 lo = root->left == left ? NULL : root->left;
1114 SPLAY_RIGHT_STEP(lo, y, left, root, true);
1120 vm_map_entry_swap(vm_map_entry_t *a, vm_map_entry_t *b)
1130 * Walk back up the two spines, flip the pointers and set max_free. The
1131 * subtrees of the root go at the bottom of llist and rlist.
1134 vm_map_splay_merge_left_walk(vm_map_entry_t header, vm_map_entry_t root,
1135 vm_map_entry_t tail, vm_size_t max_free, vm_map_entry_t llist)
1139 * The max_free values of the children of llist are in
1140 * llist->max_free and max_free. Update with the
1143 llist->max_free = max_free =
1144 vm_size_max(llist->max_free, max_free);
1145 vm_map_entry_swap(&llist->right, &tail);
1146 vm_map_entry_swap(&tail, &llist);
1147 } while (llist != header);
1153 * When llist is known to be the predecessor of root.
1155 static inline vm_size_t
1156 vm_map_splay_merge_pred(vm_map_entry_t header, vm_map_entry_t root,
1157 vm_map_entry_t llist)
1161 max_free = root->start - llist->end;
1162 if (llist != header) {
1163 max_free = vm_map_splay_merge_left_walk(header, root,
1164 root, max_free, llist);
1166 root->left = header;
1167 header->right = root;
1173 * When llist may or may not be the predecessor of root.
1175 static inline vm_size_t
1176 vm_map_splay_merge_left(vm_map_entry_t header, vm_map_entry_t root,
1177 vm_map_entry_t llist)
1181 max_free = vm_map_entry_max_free_left(root, llist);
1182 if (llist != header) {
1183 max_free = vm_map_splay_merge_left_walk(header, root,
1184 root->left == llist ? root : root->left,
1191 vm_map_splay_merge_right_walk(vm_map_entry_t header, vm_map_entry_t root,
1192 vm_map_entry_t tail, vm_size_t max_free, vm_map_entry_t rlist)
1196 * The max_free values of the children of rlist are in
1197 * rlist->max_free and max_free. Update with the
1200 rlist->max_free = max_free =
1201 vm_size_max(rlist->max_free, max_free);
1202 vm_map_entry_swap(&rlist->left, &tail);
1203 vm_map_entry_swap(&tail, &rlist);
1204 } while (rlist != header);
1210 * When rlist is known to be the succecessor of root.
1212 static inline vm_size_t
1213 vm_map_splay_merge_succ(vm_map_entry_t header, vm_map_entry_t root,
1214 vm_map_entry_t rlist)
1218 max_free = rlist->start - root->end;
1219 if (rlist != header) {
1220 max_free = vm_map_splay_merge_right_walk(header, root,
1221 root, max_free, rlist);
1223 root->right = header;
1224 header->left = root;
1230 * When rlist may or may not be the succecessor of root.
1232 static inline vm_size_t
1233 vm_map_splay_merge_right(vm_map_entry_t header, vm_map_entry_t root,
1234 vm_map_entry_t rlist)
1238 max_free = vm_map_entry_max_free_right(root, rlist);
1239 if (rlist != header) {
1240 max_free = vm_map_splay_merge_right_walk(header, root,
1241 root->right == rlist ? root : root->right,
1250 * The Sleator and Tarjan top-down splay algorithm with the
1251 * following variation. Max_free must be computed bottom-up, so
1252 * on the downward pass, maintain the left and right spines in
1253 * reverse order. Then, make a second pass up each side to fix
1254 * the pointers and compute max_free. The time bound is O(log n)
1257 * The tree is threaded, which means that there are no null pointers.
1258 * When a node has no left child, its left pointer points to its
1259 * predecessor, which the last ancestor on the search path from the root
1260 * where the search branched right. Likewise, when a node has no right
1261 * child, its right pointer points to its successor. The map header node
1262 * is the predecessor of the first map entry, and the successor of the
1265 * The new root is the vm_map_entry containing "addr", or else an
1266 * adjacent entry (lower if possible) if addr is not in the tree.
1268 * The map must be locked, and leaves it so.
1270 * Returns: the new root.
1272 static vm_map_entry_t
1273 vm_map_splay(vm_map_t map, vm_offset_t addr)
1275 vm_map_entry_t header, llist, rlist, root;
1276 vm_size_t max_free_left, max_free_right;
1278 header = &map->header;
1279 root = vm_map_splay_split(map, addr, 0, &llist, &rlist);
1281 max_free_left = vm_map_splay_merge_left(header, root, llist);
1282 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1283 } else if (llist != header) {
1285 * Recover the greatest node in the left
1286 * subtree and make it the root.
1289 llist = root->right;
1290 max_free_left = vm_map_splay_merge_left(header, root, llist);
1291 max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1292 } else if (rlist != header) {
1294 * Recover the least node in the right
1295 * subtree and make it the root.
1299 max_free_left = vm_map_splay_merge_pred(header, root, llist);
1300 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1302 /* There is no root. */
1305 root->max_free = vm_size_max(max_free_left, max_free_right);
1307 VM_MAP_ASSERT_CONSISTENT(map);
1312 * vm_map_entry_{un,}link:
1314 * Insert/remove entries from maps. On linking, if new entry clips
1315 * existing entry, trim existing entry to avoid overlap, and manage
1316 * offsets. On unlinking, merge disappearing entry with neighbor, if
1317 * called for, and manage offsets. Callers should not modify fields in
1318 * entries already mapped.
1321 vm_map_entry_link(vm_map_t map, vm_map_entry_t entry)
1323 vm_map_entry_t header, llist, rlist, root;
1324 vm_size_t max_free_left, max_free_right;
1327 "vm_map_entry_link: map %p, nentries %d, entry %p", map,
1328 map->nentries, entry);
1329 VM_MAP_ASSERT_LOCKED(map);
1331 header = &map->header;
1332 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1335 * The new entry does not overlap any existing entry in the
1336 * map, so it becomes the new root of the map tree.
1338 max_free_left = vm_map_splay_merge_pred(header, entry, llist);
1339 max_free_right = vm_map_splay_merge_succ(header, entry, rlist);
1340 } else if (entry->start == root->start) {
1342 * The new entry is a clone of root, with only the end field
1343 * changed. The root entry will be shrunk to abut the new
1344 * entry, and will be the right child of the new root entry in
1347 KASSERT(entry->end < root->end,
1348 ("%s: clip_start not within entry", __func__));
1349 vm_map_splay_findprev(root, &llist);
1350 root->offset += entry->end - root->start;
1351 root->start = entry->end;
1352 max_free_left = vm_map_splay_merge_pred(header, entry, llist);
1353 max_free_right = root->max_free = vm_size_max(
1354 vm_map_splay_merge_pred(entry, root, entry),
1355 vm_map_splay_merge_right(header, root, rlist));
1358 * The new entry is a clone of root, with only the start field
1359 * changed. The root entry will be shrunk to abut the new
1360 * entry, and will be the left child of the new root entry in
1363 KASSERT(entry->end == root->end,
1364 ("%s: clip_start not within entry", __func__));
1365 vm_map_splay_findnext(root, &rlist);
1366 entry->offset += entry->start - root->start;
1367 root->end = entry->start;
1368 max_free_left = root->max_free = vm_size_max(
1369 vm_map_splay_merge_left(header, root, llist),
1370 vm_map_splay_merge_succ(entry, root, entry));
1371 max_free_right = vm_map_splay_merge_succ(header, entry, rlist);
1373 entry->max_free = vm_size_max(max_free_left, max_free_right);
1375 VM_MAP_ASSERT_CONSISTENT(map);
1378 enum unlink_merge_type {
1384 vm_map_entry_unlink(vm_map_t map, vm_map_entry_t entry,
1385 enum unlink_merge_type op)
1387 vm_map_entry_t header, llist, rlist, root;
1388 vm_size_t max_free_left, max_free_right;
1390 VM_MAP_ASSERT_LOCKED(map);
1391 header = &map->header;
1392 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1393 KASSERT(root != NULL,
1394 ("vm_map_entry_unlink: unlink object not mapped"));
1396 vm_map_splay_findprev(root, &llist);
1397 vm_map_splay_findnext(root, &rlist);
1398 if (op == UNLINK_MERGE_NEXT) {
1399 rlist->start = root->start;
1400 rlist->offset = root->offset;
1402 if (llist != header) {
1404 llist = root->right;
1405 max_free_left = vm_map_splay_merge_left(header, root, llist);
1406 max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1407 } else if (rlist != header) {
1410 max_free_left = vm_map_splay_merge_pred(header, root, llist);
1411 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1413 header->left = header->right = header;
1417 root->max_free = vm_size_max(max_free_left, max_free_right);
1419 VM_MAP_ASSERT_CONSISTENT(map);
1421 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1422 map->nentries, entry);
1426 * vm_map_entry_resize:
1428 * Resize a vm_map_entry, recompute the amount of free space that
1429 * follows it and propagate that value up the tree.
1431 * The map must be locked, and leaves it so.
1434 vm_map_entry_resize(vm_map_t map, vm_map_entry_t entry, vm_size_t grow_amount)
1436 vm_map_entry_t header, llist, rlist, root;
1438 VM_MAP_ASSERT_LOCKED(map);
1439 header = &map->header;
1440 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1441 KASSERT(root != NULL, ("%s: resize object not mapped", __func__));
1442 vm_map_splay_findnext(root, &rlist);
1443 entry->end += grow_amount;
1444 root->max_free = vm_size_max(
1445 vm_map_splay_merge_left(header, root, llist),
1446 vm_map_splay_merge_succ(header, root, rlist));
1448 VM_MAP_ASSERT_CONSISTENT(map);
1449 CTR4(KTR_VM, "%s: map %p, nentries %d, entry %p",
1450 __func__, map, map->nentries, entry);
1454 * vm_map_lookup_entry: [ internal use only ]
1456 * Finds the map entry containing (or
1457 * immediately preceding) the specified address
1458 * in the given map; the entry is returned
1459 * in the "entry" parameter. The boolean
1460 * result indicates whether the address is
1461 * actually contained in the map.
1464 vm_map_lookup_entry(
1466 vm_offset_t address,
1467 vm_map_entry_t *entry) /* OUT */
1469 vm_map_entry_t cur, header, lbound, ubound;
1473 * If the map is empty, then the map entry immediately preceding
1474 * "address" is the map's header.
1476 header = &map->header;
1482 if (address >= cur->start && cur->end > address) {
1486 if ((locked = vm_map_locked(map)) ||
1487 sx_try_upgrade(&map->lock)) {
1489 * Splay requires a write lock on the map. However, it only
1490 * restructures the binary search tree; it does not otherwise
1491 * change the map. Thus, the map's timestamp need not change
1492 * on a temporary upgrade.
1494 cur = vm_map_splay(map, address);
1496 VM_MAP_UNLOCK_CONSISTENT(map);
1497 sx_downgrade(&map->lock);
1501 * If "address" is contained within a map entry, the new root
1502 * is that map entry. Otherwise, the new root is a map entry
1503 * immediately before or after "address".
1505 if (address < cur->start) {
1510 return (address < cur->end);
1513 * Since the map is only locked for read access, perform a
1514 * standard binary search tree lookup for "address".
1516 lbound = ubound = header;
1518 if (address < cur->start) {
1523 } else if (cur->end <= address) {
1540 * Inserts the given whole VM object into the target
1541 * map at the specified address range. The object's
1542 * size should match that of the address range.
1544 * Requires that the map be locked, and leaves it so.
1546 * If object is non-NULL, ref count must be bumped by caller
1547 * prior to making call to account for the new entry.
1550 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1551 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1553 vm_map_entry_t new_entry, next_entry, prev_entry;
1555 vm_eflags_t protoeflags;
1556 vm_inherit_t inheritance;
1558 VM_MAP_ASSERT_LOCKED(map);
1559 KASSERT(object != kernel_object ||
1560 (cow & MAP_COPY_ON_WRITE) == 0,
1561 ("vm_map_insert: kernel object and COW"));
1562 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1563 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1564 KASSERT((prot & ~max) == 0,
1565 ("prot %#x is not subset of max_prot %#x", prot, max));
1568 * Check that the start and end points are not bogus.
1570 if (start == end || !vm_map_range_valid(map, start, end))
1571 return (KERN_INVALID_ADDRESS);
1574 * Find the entry prior to the proposed starting address; if it's part
1575 * of an existing entry, this range is bogus.
1577 if (vm_map_lookup_entry(map, start, &prev_entry))
1578 return (KERN_NO_SPACE);
1581 * Assert that the next entry doesn't overlap the end point.
1583 next_entry = vm_map_entry_succ(prev_entry);
1584 if (next_entry->start < end)
1585 return (KERN_NO_SPACE);
1587 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1588 max != VM_PROT_NONE))
1589 return (KERN_INVALID_ARGUMENT);
1592 if (cow & MAP_COPY_ON_WRITE)
1593 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1594 if (cow & MAP_NOFAULT)
1595 protoeflags |= MAP_ENTRY_NOFAULT;
1596 if (cow & MAP_DISABLE_SYNCER)
1597 protoeflags |= MAP_ENTRY_NOSYNC;
1598 if (cow & MAP_DISABLE_COREDUMP)
1599 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1600 if (cow & MAP_STACK_GROWS_DOWN)
1601 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1602 if (cow & MAP_STACK_GROWS_UP)
1603 protoeflags |= MAP_ENTRY_GROWS_UP;
1604 if (cow & MAP_WRITECOUNT)
1605 protoeflags |= MAP_ENTRY_WRITECNT;
1606 if (cow & MAP_VN_EXEC)
1607 protoeflags |= MAP_ENTRY_VN_EXEC;
1608 if ((cow & MAP_CREATE_GUARD) != 0)
1609 protoeflags |= MAP_ENTRY_GUARD;
1610 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1611 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1612 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1613 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1614 if (cow & MAP_INHERIT_SHARE)
1615 inheritance = VM_INHERIT_SHARE;
1617 inheritance = VM_INHERIT_DEFAULT;
1620 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1622 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1623 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1624 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1625 return (KERN_RESOURCE_SHORTAGE);
1626 KASSERT(object == NULL ||
1627 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1628 object->cred == NULL,
1629 ("overcommit: vm_map_insert o %p", object));
1630 cred = curthread->td_ucred;
1634 /* Expand the kernel pmap, if necessary. */
1635 if (map == kernel_map && end > kernel_vm_end)
1636 pmap_growkernel(end);
1637 if (object != NULL) {
1639 * OBJ_ONEMAPPING must be cleared unless this mapping
1640 * is trivially proven to be the only mapping for any
1641 * of the object's pages. (Object granularity
1642 * reference counting is insufficient to recognize
1643 * aliases with precision.)
1645 if ((object->flags & OBJ_ANON) != 0) {
1646 VM_OBJECT_WLOCK(object);
1647 if (object->ref_count > 1 || object->shadow_count != 0)
1648 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1649 VM_OBJECT_WUNLOCK(object);
1651 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1653 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP |
1654 MAP_VN_EXEC)) == 0 &&
1655 prev_entry->end == start && (prev_entry->cred == cred ||
1656 (prev_entry->object.vm_object != NULL &&
1657 prev_entry->object.vm_object->cred == cred)) &&
1658 vm_object_coalesce(prev_entry->object.vm_object,
1660 (vm_size_t)(prev_entry->end - prev_entry->start),
1661 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1662 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1664 * We were able to extend the object. Determine if we
1665 * can extend the previous map entry to include the
1666 * new range as well.
1668 if (prev_entry->inheritance == inheritance &&
1669 prev_entry->protection == prot &&
1670 prev_entry->max_protection == max &&
1671 prev_entry->wired_count == 0) {
1672 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1673 0, ("prev_entry %p has incoherent wiring",
1675 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1676 map->size += end - prev_entry->end;
1677 vm_map_entry_resize(map, prev_entry,
1678 end - prev_entry->end);
1679 vm_map_try_merge_entries(map, prev_entry, next_entry);
1680 return (KERN_SUCCESS);
1684 * If we can extend the object but cannot extend the
1685 * map entry, we have to create a new map entry. We
1686 * must bump the ref count on the extended object to
1687 * account for it. object may be NULL.
1689 object = prev_entry->object.vm_object;
1690 offset = prev_entry->offset +
1691 (prev_entry->end - prev_entry->start);
1692 vm_object_reference(object);
1693 if (cred != NULL && object != NULL && object->cred != NULL &&
1694 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1695 /* Object already accounts for this uid. */
1703 * Create a new entry
1705 new_entry = vm_map_entry_create(map);
1706 new_entry->start = start;
1707 new_entry->end = end;
1708 new_entry->cred = NULL;
1710 new_entry->eflags = protoeflags;
1711 new_entry->object.vm_object = object;
1712 new_entry->offset = offset;
1714 new_entry->inheritance = inheritance;
1715 new_entry->protection = prot;
1716 new_entry->max_protection = max;
1717 new_entry->wired_count = 0;
1718 new_entry->wiring_thread = NULL;
1719 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1720 new_entry->next_read = start;
1722 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1723 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1724 new_entry->cred = cred;
1727 * Insert the new entry into the list
1729 vm_map_entry_link(map, new_entry);
1730 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1731 map->size += new_entry->end - new_entry->start;
1734 * Try to coalesce the new entry with both the previous and next
1735 * entries in the list. Previously, we only attempted to coalesce
1736 * with the previous entry when object is NULL. Here, we handle the
1737 * other cases, which are less common.
1739 vm_map_try_merge_entries(map, prev_entry, new_entry);
1740 vm_map_try_merge_entries(map, new_entry, next_entry);
1742 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1743 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1744 end - start, cow & MAP_PREFAULT_PARTIAL);
1747 return (KERN_SUCCESS);
1753 * Find the first fit (lowest VM address) for "length" free bytes
1754 * beginning at address >= start in the given map.
1756 * In a vm_map_entry, "max_free" is the maximum amount of
1757 * contiguous free space between an entry in its subtree and a
1758 * neighbor of that entry. This allows finding a free region in
1759 * one path down the tree, so O(log n) amortized with splay
1762 * The map must be locked, and leaves it so.
1764 * Returns: starting address if sufficient space,
1765 * vm_map_max(map)-length+1 if insufficient space.
1768 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length)
1770 vm_map_entry_t header, llist, rlist, root, y;
1771 vm_size_t left_length, max_free_left, max_free_right;
1772 vm_offset_t gap_end;
1775 * Request must fit within min/max VM address and must avoid
1778 start = MAX(start, vm_map_min(map));
1779 if (start >= vm_map_max(map) || length > vm_map_max(map) - start)
1780 return (vm_map_max(map) - length + 1);
1782 /* Empty tree means wide open address space. */
1783 if (map->root == NULL)
1787 * After splay_split, if start is within an entry, push it to the start
1788 * of the following gap. If rlist is at the end of the gap containing
1789 * start, save the end of that gap in gap_end to see if the gap is big
1790 * enough; otherwise set gap_end to start skip gap-checking and move
1791 * directly to a search of the right subtree.
1793 header = &map->header;
1794 root = vm_map_splay_split(map, start, length, &llist, &rlist);
1795 gap_end = rlist->start;
1798 if (root->right != rlist)
1800 max_free_left = vm_map_splay_merge_left(header, root, llist);
1801 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1802 } else if (rlist != header) {
1805 max_free_left = vm_map_splay_merge_pred(header, root, llist);
1806 max_free_right = vm_map_splay_merge_right(header, root, rlist);
1809 llist = root->right;
1810 max_free_left = vm_map_splay_merge_left(header, root, llist);
1811 max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1813 root->max_free = vm_size_max(max_free_left, max_free_right);
1815 VM_MAP_ASSERT_CONSISTENT(map);
1816 if (length <= gap_end - start)
1819 /* With max_free, can immediately tell if no solution. */
1820 if (root->right == header || length > root->right->max_free)
1821 return (vm_map_max(map) - length + 1);
1824 * Splay for the least large-enough gap in the right subtree.
1826 llist = rlist = header;
1827 for (left_length = 0;;
1828 left_length = vm_map_entry_max_free_left(root, llist)) {
1829 if (length <= left_length)
1830 SPLAY_LEFT_STEP(root, y, llist, rlist,
1831 length <= vm_map_entry_max_free_left(y, llist));
1833 SPLAY_RIGHT_STEP(root, y, llist, rlist,
1834 length > vm_map_entry_max_free_left(y, root));
1839 llist = root->right;
1840 max_free_left = vm_map_splay_merge_left(header, root, llist);
1841 if (rlist == header) {
1842 root->max_free = vm_size_max(max_free_left,
1843 vm_map_splay_merge_succ(header, root, rlist));
1847 y->max_free = vm_size_max(
1848 vm_map_splay_merge_pred(root, y, root),
1849 vm_map_splay_merge_right(header, y, rlist));
1850 root->max_free = vm_size_max(max_free_left, y->max_free);
1853 VM_MAP_ASSERT_CONSISTENT(map);
1858 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1859 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1860 vm_prot_t max, int cow)
1865 end = start + length;
1866 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1868 ("vm_map_fixed: non-NULL backing object for stack"));
1870 VM_MAP_RANGE_CHECK(map, start, end);
1871 if ((cow & MAP_CHECK_EXCL) == 0)
1872 vm_map_delete(map, start, end);
1873 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1874 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1877 result = vm_map_insert(map, object, offset, start, end,
1884 static const int aslr_pages_rnd_64[2] = {0x1000, 0x10};
1885 static const int aslr_pages_rnd_32[2] = {0x100, 0x4};
1887 static int cluster_anon = 1;
1888 SYSCTL_INT(_vm, OID_AUTO, cluster_anon, CTLFLAG_RW,
1890 "Cluster anonymous mappings: 0 = no, 1 = yes if no hint, 2 = always");
1893 clustering_anon_allowed(vm_offset_t addr)
1896 switch (cluster_anon) {
1907 static long aslr_restarts;
1908 SYSCTL_LONG(_vm, OID_AUTO, aslr_restarts, CTLFLAG_RD,
1910 "Number of aslr failures");
1912 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
1915 * Searches for the specified amount of free space in the given map with the
1916 * specified alignment. Performs an address-ordered, first-fit search from
1917 * the given address "*addr", with an optional upper bound "max_addr". If the
1918 * parameter "alignment" is zero, then the alignment is computed from the
1919 * given (object, offset) pair so as to enable the greatest possible use of
1920 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1921 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1923 * The map must be locked. Initially, there must be at least "length" bytes
1924 * of free space at the given address.
1927 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1928 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1929 vm_offset_t alignment)
1931 vm_offset_t aligned_addr, free_addr;
1933 VM_MAP_ASSERT_LOCKED(map);
1935 KASSERT(free_addr == vm_map_findspace(map, free_addr, length),
1936 ("caller failed to provide space %#jx at address %p",
1937 (uintmax_t)length, (void *)free_addr));
1940 * At the start of every iteration, the free space at address
1941 * "*addr" is at least "length" bytes.
1944 pmap_align_superpage(object, offset, addr, length);
1945 else if ((*addr & (alignment - 1)) != 0) {
1946 *addr &= ~(alignment - 1);
1949 aligned_addr = *addr;
1950 if (aligned_addr == free_addr) {
1952 * Alignment did not change "*addr", so "*addr" must
1953 * still provide sufficient free space.
1955 return (KERN_SUCCESS);
1959 * Test for address wrap on "*addr". A wrapped "*addr" could
1960 * be a valid address, in which case vm_map_findspace() cannot
1961 * be relied upon to fail.
1963 if (aligned_addr < free_addr)
1964 return (KERN_NO_SPACE);
1965 *addr = vm_map_findspace(map, aligned_addr, length);
1966 if (*addr + length > vm_map_max(map) ||
1967 (max_addr != 0 && *addr + length > max_addr))
1968 return (KERN_NO_SPACE);
1970 if (free_addr == aligned_addr) {
1972 * If a successful call to vm_map_findspace() did not
1973 * change "*addr", then "*addr" must still be aligned
1974 * and provide sufficient free space.
1976 return (KERN_SUCCESS);
1982 * vm_map_find finds an unallocated region in the target address
1983 * map with the given length. The search is defined to be
1984 * first-fit from the specified address; the region found is
1985 * returned in the same parameter.
1987 * If object is non-NULL, ref count must be bumped by caller
1988 * prior to making call to account for the new entry.
1991 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1992 vm_offset_t *addr, /* IN/OUT */
1993 vm_size_t length, vm_offset_t max_addr, int find_space,
1994 vm_prot_t prot, vm_prot_t max, int cow)
1996 vm_offset_t alignment, curr_min_addr, min_addr;
1997 int gap, pidx, rv, try;
1998 bool cluster, en_aslr, update_anon;
2000 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
2002 ("vm_map_find: non-NULL backing object for stack"));
2003 MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE &&
2004 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0));
2005 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
2006 (object->flags & OBJ_COLORED) == 0))
2007 find_space = VMFS_ANY_SPACE;
2008 if (find_space >> 8 != 0) {
2009 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
2010 alignment = (vm_offset_t)1 << (find_space >> 8);
2013 en_aslr = (map->flags & MAP_ASLR) != 0;
2014 update_anon = cluster = clustering_anon_allowed(*addr) &&
2015 (map->flags & MAP_IS_SUB_MAP) == 0 && max_addr == 0 &&
2016 find_space != VMFS_NO_SPACE && object == NULL &&
2017 (cow & (MAP_INHERIT_SHARE | MAP_STACK_GROWS_UP |
2018 MAP_STACK_GROWS_DOWN)) == 0 && prot != PROT_NONE;
2019 curr_min_addr = min_addr = *addr;
2020 if (en_aslr && min_addr == 0 && !cluster &&
2021 find_space != VMFS_NO_SPACE &&
2022 (map->flags & MAP_ASLR_IGNSTART) != 0)
2023 curr_min_addr = min_addr = vm_map_min(map);
2027 curr_min_addr = map->anon_loc;
2028 if (curr_min_addr == 0)
2031 if (find_space != VMFS_NO_SPACE) {
2032 KASSERT(find_space == VMFS_ANY_SPACE ||
2033 find_space == VMFS_OPTIMAL_SPACE ||
2034 find_space == VMFS_SUPER_SPACE ||
2035 alignment != 0, ("unexpected VMFS flag"));
2038 * When creating an anonymous mapping, try clustering
2039 * with an existing anonymous mapping first.
2041 * We make up to two attempts to find address space
2042 * for a given find_space value. The first attempt may
2043 * apply randomization or may cluster with an existing
2044 * anonymous mapping. If this first attempt fails,
2045 * perform a first-fit search of the available address
2048 * If all tries failed, and find_space is
2049 * VMFS_OPTIMAL_SPACE, fallback to VMFS_ANY_SPACE.
2050 * Again enable clustering and randomization.
2057 * Second try: we failed either to find a
2058 * suitable region for randomizing the
2059 * allocation, or to cluster with an existing
2060 * mapping. Retry with free run.
2062 curr_min_addr = (map->flags & MAP_ASLR_IGNSTART) != 0 ?
2063 vm_map_min(map) : min_addr;
2064 atomic_add_long(&aslr_restarts, 1);
2067 if (try == 1 && en_aslr && !cluster) {
2069 * Find space for allocation, including
2070 * gap needed for later randomization.
2072 pidx = MAXPAGESIZES > 1 && pagesizes[1] != 0 &&
2073 (find_space == VMFS_SUPER_SPACE || find_space ==
2074 VMFS_OPTIMAL_SPACE) ? 1 : 0;
2075 gap = vm_map_max(map) > MAP_32BIT_MAX_ADDR &&
2076 (max_addr == 0 || max_addr > MAP_32BIT_MAX_ADDR) ?
2077 aslr_pages_rnd_64[pidx] : aslr_pages_rnd_32[pidx];
2078 *addr = vm_map_findspace(map, curr_min_addr,
2079 length + gap * pagesizes[pidx]);
2080 if (*addr + length + gap * pagesizes[pidx] >
2083 /* And randomize the start address. */
2084 *addr += (arc4random() % gap) * pagesizes[pidx];
2085 if (max_addr != 0 && *addr + length > max_addr)
2088 *addr = vm_map_findspace(map, curr_min_addr, length);
2089 if (*addr + length > vm_map_max(map) ||
2090 (max_addr != 0 && *addr + length > max_addr)) {
2101 if (find_space != VMFS_ANY_SPACE &&
2102 (rv = vm_map_alignspace(map, object, offset, addr, length,
2103 max_addr, alignment)) != KERN_SUCCESS) {
2104 if (find_space == VMFS_OPTIMAL_SPACE) {
2105 find_space = VMFS_ANY_SPACE;
2106 curr_min_addr = min_addr;
2107 cluster = update_anon;
2113 } else if ((cow & MAP_REMAP) != 0) {
2114 if (!vm_map_range_valid(map, *addr, *addr + length)) {
2115 rv = KERN_INVALID_ADDRESS;
2118 vm_map_delete(map, *addr, *addr + length);
2120 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
2121 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
2124 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
2127 if (rv == KERN_SUCCESS && update_anon)
2128 map->anon_loc = *addr + length;
2135 * vm_map_find_min() is a variant of vm_map_find() that takes an
2136 * additional parameter (min_addr) and treats the given address
2137 * (*addr) differently. Specifically, it treats *addr as a hint
2138 * and not as the minimum address where the mapping is created.
2140 * This function works in two phases. First, it tries to
2141 * allocate above the hint. If that fails and the hint is
2142 * greater than min_addr, it performs a second pass, replacing
2143 * the hint with min_addr as the minimum address for the
2147 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
2148 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
2149 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
2157 rv = vm_map_find(map, object, offset, addr, length, max_addr,
2158 find_space, prot, max, cow);
2159 if (rv == KERN_SUCCESS || min_addr >= hint)
2161 *addr = hint = min_addr;
2166 * A map entry with any of the following flags set must not be merged with
2169 #define MAP_ENTRY_NOMERGE_MASK (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \
2170 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_VN_EXEC)
2173 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
2176 KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 ||
2177 (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0,
2178 ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable",
2180 return (prev->end == entry->start &&
2181 prev->object.vm_object == entry->object.vm_object &&
2182 (prev->object.vm_object == NULL ||
2183 prev->offset + (prev->end - prev->start) == entry->offset) &&
2184 prev->eflags == entry->eflags &&
2185 prev->protection == entry->protection &&
2186 prev->max_protection == entry->max_protection &&
2187 prev->inheritance == entry->inheritance &&
2188 prev->wired_count == entry->wired_count &&
2189 prev->cred == entry->cred);
2193 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
2197 * If the backing object is a vnode object, vm_object_deallocate()
2198 * calls vrele(). However, vrele() does not lock the vnode because
2199 * the vnode has additional references. Thus, the map lock can be
2200 * kept without causing a lock-order reversal with the vnode lock.
2202 * Since we count the number of virtual page mappings in
2203 * object->un_pager.vnp.writemappings, the writemappings value
2204 * should not be adjusted when the entry is disposed of.
2206 if (entry->object.vm_object != NULL)
2207 vm_object_deallocate(entry->object.vm_object);
2208 if (entry->cred != NULL)
2209 crfree(entry->cred);
2210 vm_map_entry_dispose(map, entry);
2214 * vm_map_try_merge_entries:
2216 * Compare the given map entry to its predecessor, and merge its precessor
2217 * into it if possible. The entry remains valid, and may be extended.
2218 * The predecessor may be deleted.
2220 * The map must be locked.
2223 vm_map_try_merge_entries(vm_map_t map, vm_map_entry_t prev_entry,
2224 vm_map_entry_t entry)
2227 VM_MAP_ASSERT_LOCKED(map);
2228 if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 &&
2229 vm_map_mergeable_neighbors(prev_entry, entry)) {
2230 vm_map_entry_unlink(map, prev_entry, UNLINK_MERGE_NEXT);
2231 vm_map_merged_neighbor_dispose(map, prev_entry);
2236 * vm_map_entry_back:
2238 * Allocate an object to back a map entry.
2241 vm_map_entry_back(vm_map_entry_t entry)
2245 KASSERT(entry->object.vm_object == NULL,
2246 ("map entry %p has backing object", entry));
2247 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2248 ("map entry %p is a submap", entry));
2249 object = vm_object_allocate_anon(atop(entry->end - entry->start), NULL,
2250 entry->cred, entry->end - entry->start);
2251 entry->object.vm_object = object;
2257 * vm_map_entry_charge_object
2259 * If there is no object backing this entry, create one. Otherwise, if
2260 * the entry has cred, give it to the backing object.
2263 vm_map_entry_charge_object(vm_map_t map, vm_map_entry_t entry)
2266 VM_MAP_ASSERT_LOCKED(map);
2267 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2268 ("map entry %p is a submap", entry));
2269 if (entry->object.vm_object == NULL && !map->system_map &&
2270 (entry->eflags & MAP_ENTRY_GUARD) == 0)
2271 vm_map_entry_back(entry);
2272 else if (entry->object.vm_object != NULL &&
2273 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
2274 entry->cred != NULL) {
2275 VM_OBJECT_WLOCK(entry->object.vm_object);
2276 KASSERT(entry->object.vm_object->cred == NULL,
2277 ("OVERCOMMIT: %s: both cred e %p", __func__, entry));
2278 entry->object.vm_object->cred = entry->cred;
2279 entry->object.vm_object->charge = entry->end - entry->start;
2280 VM_OBJECT_WUNLOCK(entry->object.vm_object);
2286 * vm_map_entry_clone
2288 * Create a duplicate map entry for clipping.
2290 static vm_map_entry_t
2291 vm_map_entry_clone(vm_map_t map, vm_map_entry_t entry)
2293 vm_map_entry_t new_entry;
2295 VM_MAP_ASSERT_LOCKED(map);
2298 * Create a backing object now, if none exists, so that more individual
2299 * objects won't be created after the map entry is split.
2301 vm_map_entry_charge_object(map, entry);
2303 /* Clone the entry. */
2304 new_entry = vm_map_entry_create(map);
2305 *new_entry = *entry;
2306 if (new_entry->cred != NULL)
2307 crhold(entry->cred);
2308 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2309 vm_object_reference(new_entry->object.vm_object);
2310 vm_map_entry_set_vnode_text(new_entry, true);
2312 * The object->un_pager.vnp.writemappings for the object of
2313 * MAP_ENTRY_WRITECNT type entry shall be kept as is here. The
2314 * virtual pages are re-distributed among the clipped entries,
2315 * so the sum is left the same.
2322 * vm_map_clip_start: [ internal use only ]
2324 * Asserts that the given entry begins at or after
2325 * the specified address; if necessary,
2326 * it splits the entry into two.
2329 vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
2331 vm_map_entry_t new_entry;
2333 if (!map->system_map)
2334 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2335 "%s: map %p entry %p start 0x%jx", __func__, map, entry,
2338 if (start <= entry->start)
2341 VM_MAP_ASSERT_LOCKED(map);
2342 KASSERT(entry->end > start && entry->start < start,
2343 ("%s: invalid clip of entry %p", __func__, entry));
2345 new_entry = vm_map_entry_clone(map, entry);
2348 * Split off the front portion. Insert the new entry BEFORE this one,
2349 * so that this entry has the specified starting address.
2351 new_entry->end = start;
2352 vm_map_entry_link(map, new_entry);
2356 * vm_map_lookup_clip_start:
2358 * Find the entry at or just after 'start', and clip it if 'start' is in
2359 * the interior of the entry. Return entry after 'start', and in
2360 * prev_entry set the entry before 'start'.
2362 static inline vm_map_entry_t
2363 vm_map_lookup_clip_start(vm_map_t map, vm_offset_t start,
2364 vm_map_entry_t *prev_entry)
2366 vm_map_entry_t entry;
2368 if (!map->system_map)
2369 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2370 "%s: map %p start 0x%jx prev %p", __func__, map,
2371 (uintmax_t)start, prev_entry);
2373 if (vm_map_lookup_entry(map, start, prev_entry)) {
2374 entry = *prev_entry;
2375 vm_map_clip_start(map, entry, start);
2376 *prev_entry = vm_map_entry_pred(entry);
2378 entry = vm_map_entry_succ(*prev_entry);
2383 * vm_map_clip_end: [ internal use only ]
2385 * Asserts that the given entry ends at or before
2386 * the specified address; if necessary,
2387 * it splits the entry into two.
2390 vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
2392 vm_map_entry_t new_entry;
2394 if (!map->system_map)
2395 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2396 "%s: map %p entry %p end 0x%jx", __func__, map, entry,
2399 if (end >= entry->end)
2402 VM_MAP_ASSERT_LOCKED(map);
2403 KASSERT(entry->start < end && entry->end > end,
2404 ("%s: invalid clip of entry %p", __func__, entry));
2406 new_entry = vm_map_entry_clone(map, entry);
2409 * Split off the back portion. Insert the new entry AFTER this one,
2410 * so that this entry has the specified ending address.
2412 new_entry->start = end;
2413 vm_map_entry_link(map, new_entry);
2417 * vm_map_submap: [ kernel use only ]
2419 * Mark the given range as handled by a subordinate map.
2421 * This range must have been created with vm_map_find,
2422 * and no other operations may have been performed on this
2423 * range prior to calling vm_map_submap.
2425 * Only a limited number of operations can be performed
2426 * within this rage after calling vm_map_submap:
2428 * [Don't try vm_map_copy!]
2430 * To remove a submapping, one must first remove the
2431 * range from the superior map, and then destroy the
2432 * submap (if desired). [Better yet, don't try it.]
2441 vm_map_entry_t entry;
2444 result = KERN_INVALID_ARGUMENT;
2446 vm_map_lock(submap);
2447 submap->flags |= MAP_IS_SUB_MAP;
2448 vm_map_unlock(submap);
2451 VM_MAP_RANGE_CHECK(map, start, end);
2452 if (vm_map_lookup_entry(map, start, &entry) && entry->end >= end &&
2453 (entry->eflags & MAP_ENTRY_COW) == 0 &&
2454 entry->object.vm_object == NULL) {
2455 vm_map_clip_start(map, entry, start);
2456 vm_map_clip_end(map, entry, end);
2457 entry->object.sub_map = submap;
2458 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
2459 result = KERN_SUCCESS;
2463 if (result != KERN_SUCCESS) {
2464 vm_map_lock(submap);
2465 submap->flags &= ~MAP_IS_SUB_MAP;
2466 vm_map_unlock(submap);
2472 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
2474 #define MAX_INIT_PT 96
2477 * vm_map_pmap_enter:
2479 * Preload the specified map's pmap with mappings to the specified
2480 * object's memory-resident pages. No further physical pages are
2481 * allocated, and no further virtual pages are retrieved from secondary
2482 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
2483 * limited number of page mappings are created at the low-end of the
2484 * specified address range. (For this purpose, a superpage mapping
2485 * counts as one page mapping.) Otherwise, all resident pages within
2486 * the specified address range are mapped.
2489 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
2490 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
2493 vm_page_t p, p_start;
2494 vm_pindex_t mask, psize, threshold, tmpidx;
2496 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
2498 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2499 VM_OBJECT_WLOCK(object);
2500 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2501 pmap_object_init_pt(map->pmap, addr, object, pindex,
2503 VM_OBJECT_WUNLOCK(object);
2506 VM_OBJECT_LOCK_DOWNGRADE(object);
2508 VM_OBJECT_RLOCK(object);
2511 if (psize + pindex > object->size) {
2512 if (pindex >= object->size) {
2513 VM_OBJECT_RUNLOCK(object);
2516 psize = object->size - pindex;
2521 threshold = MAX_INIT_PT;
2523 p = vm_page_find_least(object, pindex);
2525 * Assert: the variable p is either (1) the page with the
2526 * least pindex greater than or equal to the parameter pindex
2530 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2531 p = TAILQ_NEXT(p, listq)) {
2533 * don't allow an madvise to blow away our really
2534 * free pages allocating pv entries.
2536 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2537 vm_page_count_severe()) ||
2538 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2539 tmpidx >= threshold)) {
2543 if (vm_page_all_valid(p)) {
2544 if (p_start == NULL) {
2545 start = addr + ptoa(tmpidx);
2548 /* Jump ahead if a superpage mapping is possible. */
2549 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2550 (pagesizes[p->psind] - 1)) == 0) {
2551 mask = atop(pagesizes[p->psind]) - 1;
2552 if (tmpidx + mask < psize &&
2553 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2558 } else if (p_start != NULL) {
2559 pmap_enter_object(map->pmap, start, addr +
2560 ptoa(tmpidx), p_start, prot);
2564 if (p_start != NULL)
2565 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2567 VM_OBJECT_RUNLOCK(object);
2573 * Sets the protection of the specified address
2574 * region in the target map. If "set_max" is
2575 * specified, the maximum protection is to be set;
2576 * otherwise, only the current protection is affected.
2579 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2580 vm_prot_t new_prot, boolean_t set_max)
2582 vm_map_entry_t entry, first_entry, in_tran, prev_entry;
2589 return (KERN_SUCCESS);
2596 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2597 * need to fault pages into the map and will drop the map lock while
2598 * doing so, and the VM object may end up in an inconsistent state if we
2599 * update the protection on the map entry in between faults.
2601 vm_map_wait_busy(map);
2603 VM_MAP_RANGE_CHECK(map, start, end);
2605 if (!vm_map_lookup_entry(map, start, &first_entry))
2606 first_entry = vm_map_entry_succ(first_entry);
2609 * Make a first pass to check for protection violations.
2611 for (entry = first_entry; entry->start < end;
2612 entry = vm_map_entry_succ(entry)) {
2613 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
2615 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
2617 return (KERN_INVALID_ARGUMENT);
2619 if ((new_prot & entry->max_protection) != new_prot) {
2621 return (KERN_PROTECTION_FAILURE);
2623 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0)
2628 * Postpone the operation until all in-transition map entries have
2629 * stabilized. An in-transition entry might already have its pages
2630 * wired and wired_count incremented, but not yet have its
2631 * MAP_ENTRY_USER_WIRED flag set. In which case, we would fail to call
2632 * vm_fault_copy_entry() in the final loop below.
2634 if (in_tran != NULL) {
2635 in_tran->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2636 vm_map_unlock_and_wait(map, 0);
2641 * Before changing the protections, try to reserve swap space for any
2642 * private (i.e., copy-on-write) mappings that are transitioning from
2643 * read-only to read/write access. If a reservation fails, break out
2644 * of this loop early and let the next loop simplify the entries, since
2645 * some may now be mergeable.
2648 vm_map_clip_start(map, first_entry, start);
2649 for (entry = first_entry; entry->start < end;
2650 entry = vm_map_entry_succ(entry)) {
2651 vm_map_clip_end(map, entry, end);
2654 ((new_prot & ~entry->protection) & VM_PROT_WRITE) == 0 ||
2655 ENTRY_CHARGED(entry) ||
2656 (entry->eflags & MAP_ENTRY_GUARD) != 0) {
2660 cred = curthread->td_ucred;
2661 obj = entry->object.vm_object;
2664 (entry->eflags & MAP_ENTRY_NEEDS_COPY) != 0) {
2665 if (!swap_reserve(entry->end - entry->start)) {
2666 rv = KERN_RESOURCE_SHORTAGE;
2675 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP)
2677 VM_OBJECT_WLOCK(obj);
2678 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2679 VM_OBJECT_WUNLOCK(obj);
2684 * Charge for the whole object allocation now, since
2685 * we cannot distinguish between non-charged and
2686 * charged clipped mapping of the same object later.
2688 KASSERT(obj->charge == 0,
2689 ("vm_map_protect: object %p overcharged (entry %p)",
2691 if (!swap_reserve(ptoa(obj->size))) {
2692 VM_OBJECT_WUNLOCK(obj);
2693 rv = KERN_RESOURCE_SHORTAGE;
2700 obj->charge = ptoa(obj->size);
2701 VM_OBJECT_WUNLOCK(obj);
2705 * If enough swap space was available, go back and fix up protections.
2706 * Otherwise, just simplify entries, since some may have been modified.
2707 * [Note that clipping is not necessary the second time.]
2709 for (prev_entry = vm_map_entry_pred(first_entry), entry = first_entry;
2711 vm_map_try_merge_entries(map, prev_entry, entry),
2712 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
2713 if (rv != KERN_SUCCESS ||
2714 (entry->eflags & MAP_ENTRY_GUARD) != 0)
2717 old_prot = entry->protection;
2721 (entry->max_protection = new_prot) &
2724 entry->protection = new_prot;
2727 * For user wired map entries, the normal lazy evaluation of
2728 * write access upgrades through soft page faults is
2729 * undesirable. Instead, immediately copy any pages that are
2730 * copy-on-write and enable write access in the physical map.
2732 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2733 (entry->protection & VM_PROT_WRITE) != 0 &&
2734 (old_prot & VM_PROT_WRITE) == 0)
2735 vm_fault_copy_entry(map, map, entry, entry, NULL);
2738 * When restricting access, update the physical map. Worry
2739 * about copy-on-write here.
2741 if ((old_prot & ~entry->protection) != 0) {
2742 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2744 pmap_protect(map->pmap, entry->start,
2746 entry->protection & MASK(entry));
2750 vm_map_try_merge_entries(map, prev_entry, entry);
2758 * This routine traverses a processes map handling the madvise
2759 * system call. Advisories are classified as either those effecting
2760 * the vm_map_entry structure, or those effecting the underlying
2770 vm_map_entry_t entry, prev_entry;
2774 * Some madvise calls directly modify the vm_map_entry, in which case
2775 * we need to use an exclusive lock on the map and we need to perform
2776 * various clipping operations. Otherwise we only need a read-lock
2781 case MADV_SEQUENTIAL:
2798 vm_map_lock_read(map);
2805 * Locate starting entry and clip if necessary.
2807 VM_MAP_RANGE_CHECK(map, start, end);
2811 * madvise behaviors that are implemented in the vm_map_entry.
2813 * We clip the vm_map_entry so that behavioral changes are
2814 * limited to the specified address range.
2816 for (entry = vm_map_lookup_clip_start(map, start, &prev_entry);
2818 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
2819 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
2822 vm_map_clip_end(map, entry, end);
2826 vm_map_entry_set_behavior(entry,
2827 MAP_ENTRY_BEHAV_NORMAL);
2829 case MADV_SEQUENTIAL:
2830 vm_map_entry_set_behavior(entry,
2831 MAP_ENTRY_BEHAV_SEQUENTIAL);
2834 vm_map_entry_set_behavior(entry,
2835 MAP_ENTRY_BEHAV_RANDOM);
2838 entry->eflags |= MAP_ENTRY_NOSYNC;
2841 entry->eflags &= ~MAP_ENTRY_NOSYNC;
2844 entry->eflags |= MAP_ENTRY_NOCOREDUMP;
2847 entry->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2852 vm_map_try_merge_entries(map, prev_entry, entry);
2854 vm_map_try_merge_entries(map, prev_entry, entry);
2857 vm_pindex_t pstart, pend;
2860 * madvise behaviors that are implemented in the underlying
2863 * Since we don't clip the vm_map_entry, we have to clip
2864 * the vm_object pindex and count.
2866 if (!vm_map_lookup_entry(map, start, &entry))
2867 entry = vm_map_entry_succ(entry);
2868 for (; entry->start < end;
2869 entry = vm_map_entry_succ(entry)) {
2870 vm_offset_t useEnd, useStart;
2872 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
2876 * MADV_FREE would otherwise rewind time to
2877 * the creation of the shadow object. Because
2878 * we hold the VM map read-locked, neither the
2879 * entry's object nor the presence of a
2880 * backing object can change.
2882 if (behav == MADV_FREE &&
2883 entry->object.vm_object != NULL &&
2884 entry->object.vm_object->backing_object != NULL)
2887 pstart = OFF_TO_IDX(entry->offset);
2888 pend = pstart + atop(entry->end - entry->start);
2889 useStart = entry->start;
2890 useEnd = entry->end;
2892 if (entry->start < start) {
2893 pstart += atop(start - entry->start);
2896 if (entry->end > end) {
2897 pend -= atop(entry->end - end);
2905 * Perform the pmap_advise() before clearing
2906 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2907 * concurrent pmap operation, such as pmap_remove(),
2908 * could clear a reference in the pmap and set
2909 * PGA_REFERENCED on the page before the pmap_advise()
2910 * had completed. Consequently, the page would appear
2911 * referenced based upon an old reference that
2912 * occurred before this pmap_advise() ran.
2914 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2915 pmap_advise(map->pmap, useStart, useEnd,
2918 vm_object_madvise(entry->object.vm_object, pstart,
2922 * Pre-populate paging structures in the
2923 * WILLNEED case. For wired entries, the
2924 * paging structures are already populated.
2926 if (behav == MADV_WILLNEED &&
2927 entry->wired_count == 0) {
2928 vm_map_pmap_enter(map,
2931 entry->object.vm_object,
2933 ptoa(pend - pstart),
2934 MAP_PREFAULT_MADVISE
2938 vm_map_unlock_read(map);
2946 * Sets the inheritance of the specified address
2947 * range in the target map. Inheritance
2948 * affects how the map will be shared with
2949 * child maps at the time of vmspace_fork.
2952 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2953 vm_inherit_t new_inheritance)
2955 vm_map_entry_t entry, prev_entry;
2957 switch (new_inheritance) {
2958 case VM_INHERIT_NONE:
2959 case VM_INHERIT_COPY:
2960 case VM_INHERIT_SHARE:
2961 case VM_INHERIT_ZERO:
2964 return (KERN_INVALID_ARGUMENT);
2967 return (KERN_SUCCESS);
2969 VM_MAP_RANGE_CHECK(map, start, end);
2970 for (entry = vm_map_lookup_clip_start(map, start, &prev_entry);
2972 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
2973 vm_map_clip_end(map, entry, end);
2974 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2975 new_inheritance != VM_INHERIT_ZERO)
2976 entry->inheritance = new_inheritance;
2977 vm_map_try_merge_entries(map, prev_entry, entry);
2979 vm_map_try_merge_entries(map, prev_entry, entry);
2981 return (KERN_SUCCESS);
2985 * vm_map_entry_in_transition:
2987 * Release the map lock, and sleep until the entry is no longer in
2988 * transition. Awake and acquire the map lock. If the map changed while
2989 * another held the lock, lookup a possibly-changed entry at or after the
2990 * 'start' position of the old entry.
2992 static vm_map_entry_t
2993 vm_map_entry_in_transition(vm_map_t map, vm_offset_t in_start,
2994 vm_offset_t *io_end, bool holes_ok, vm_map_entry_t in_entry)
2996 vm_map_entry_t entry;
2998 u_int last_timestamp;
3000 VM_MAP_ASSERT_LOCKED(map);
3001 KASSERT((in_entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3002 ("not in-tranition map entry %p", in_entry));
3004 * We have not yet clipped the entry.
3006 start = MAX(in_start, in_entry->start);
3007 in_entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3008 last_timestamp = map->timestamp;
3009 if (vm_map_unlock_and_wait(map, 0)) {
3011 * Allow interruption of user wiring/unwiring?
3015 if (last_timestamp + 1 == map->timestamp)
3019 * Look again for the entry because the map was modified while it was
3020 * unlocked. Specifically, the entry may have been clipped, merged, or
3023 if (!vm_map_lookup_entry(map, start, &entry)) {
3028 entry = vm_map_entry_succ(entry);
3036 * Implements both kernel and user unwiring.
3039 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
3042 vm_map_entry_t entry, first_entry, next_entry, prev_entry;
3044 bool holes_ok, need_wakeup, user_unwire;
3047 return (KERN_SUCCESS);
3048 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3049 user_unwire = (flags & VM_MAP_WIRE_USER) != 0;
3051 VM_MAP_RANGE_CHECK(map, start, end);
3052 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3054 first_entry = vm_map_entry_succ(first_entry);
3057 return (KERN_INVALID_ADDRESS);
3061 for (entry = first_entry; entry->start < end; entry = next_entry) {
3062 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3064 * We have not yet clipped the entry.
3066 next_entry = vm_map_entry_in_transition(map, start,
3067 &end, holes_ok, entry);
3068 if (next_entry == NULL) {
3069 if (entry == first_entry) {
3071 return (KERN_INVALID_ADDRESS);
3073 rv = KERN_INVALID_ADDRESS;
3076 first_entry = (entry == first_entry) ?
3080 vm_map_clip_start(map, entry, start);
3081 vm_map_clip_end(map, entry, end);
3083 * Mark the entry in case the map lock is released. (See
3086 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3087 entry->wiring_thread == NULL,
3088 ("owned map entry %p", entry));
3089 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3090 entry->wiring_thread = curthread;
3091 next_entry = vm_map_entry_succ(entry);
3093 * Check the map for holes in the specified region.
3094 * If holes_ok, skip this check.
3097 entry->end < end && next_entry->start > entry->end) {
3099 rv = KERN_INVALID_ADDRESS;
3103 * If system unwiring, require that the entry is system wired.
3106 vm_map_entry_system_wired_count(entry) == 0) {
3108 rv = KERN_INVALID_ARGUMENT;
3112 need_wakeup = false;
3113 if (first_entry == NULL &&
3114 !vm_map_lookup_entry(map, start, &first_entry)) {
3115 KASSERT(holes_ok, ("vm_map_unwire: lookup failed"));
3116 prev_entry = first_entry;
3117 entry = vm_map_entry_succ(first_entry);
3119 prev_entry = vm_map_entry_pred(first_entry);
3120 entry = first_entry;
3122 for (; entry->start < end;
3123 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
3125 * If holes_ok was specified, an empty
3126 * space in the unwired region could have been mapped
3127 * while the map lock was dropped for draining
3128 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
3129 * could be simultaneously wiring this new mapping
3130 * entry. Detect these cases and skip any entries
3131 * marked as in transition by us.
3133 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3134 entry->wiring_thread != curthread) {
3136 ("vm_map_unwire: !HOLESOK and new/changed entry"));
3140 if (rv == KERN_SUCCESS && (!user_unwire ||
3141 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
3142 if (entry->wired_count == 1)
3143 vm_map_entry_unwire(map, entry);
3145 entry->wired_count--;
3147 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3149 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3150 ("vm_map_unwire: in-transition flag missing %p", entry));
3151 KASSERT(entry->wiring_thread == curthread,
3152 ("vm_map_unwire: alien wire %p", entry));
3153 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
3154 entry->wiring_thread = NULL;
3155 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3156 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3159 vm_map_try_merge_entries(map, prev_entry, entry);
3161 vm_map_try_merge_entries(map, prev_entry, entry);
3169 vm_map_wire_user_count_sub(u_long npages)
3172 atomic_subtract_long(&vm_user_wire_count, npages);
3176 vm_map_wire_user_count_add(u_long npages)
3180 wired = vm_user_wire_count;
3182 if (npages + wired > vm_page_max_user_wired)
3184 } while (!atomic_fcmpset_long(&vm_user_wire_count, &wired,
3191 * vm_map_wire_entry_failure:
3193 * Handle a wiring failure on the given entry.
3195 * The map should be locked.
3198 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
3199 vm_offset_t failed_addr)
3202 VM_MAP_ASSERT_LOCKED(map);
3203 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
3204 entry->wired_count == 1,
3205 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
3206 KASSERT(failed_addr < entry->end,
3207 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
3210 * If any pages at the start of this entry were successfully wired,
3213 if (failed_addr > entry->start) {
3214 pmap_unwire(map->pmap, entry->start, failed_addr);
3215 vm_object_unwire(entry->object.vm_object, entry->offset,
3216 failed_addr - entry->start, PQ_ACTIVE);
3220 * Assign an out-of-range value to represent the failure to wire this
3223 entry->wired_count = -1;
3227 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3232 rv = vm_map_wire_locked(map, start, end, flags);
3238 * vm_map_wire_locked:
3240 * Implements both kernel and user wiring. Returns with the map locked,
3241 * the map lock may be dropped.
3244 vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3246 vm_map_entry_t entry, first_entry, next_entry, prev_entry;
3247 vm_offset_t faddr, saved_end, saved_start;
3249 u_int last_timestamp;
3251 bool holes_ok, need_wakeup, user_wire;
3254 VM_MAP_ASSERT_LOCKED(map);
3257 return (KERN_SUCCESS);
3259 if (flags & VM_MAP_WIRE_WRITE)
3260 prot |= VM_PROT_WRITE;
3261 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3262 user_wire = (flags & VM_MAP_WIRE_USER) != 0;
3263 VM_MAP_RANGE_CHECK(map, start, end);
3264 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3266 first_entry = vm_map_entry_succ(first_entry);
3268 return (KERN_INVALID_ADDRESS);
3270 for (entry = first_entry; entry->start < end; entry = next_entry) {
3271 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3273 * We have not yet clipped the entry.
3275 next_entry = vm_map_entry_in_transition(map, start,
3276 &end, holes_ok, entry);
3277 if (next_entry == NULL) {
3278 if (entry == first_entry)
3279 return (KERN_INVALID_ADDRESS);
3280 rv = KERN_INVALID_ADDRESS;
3283 first_entry = (entry == first_entry) ?
3287 vm_map_clip_start(map, entry, start);
3288 vm_map_clip_end(map, entry, end);
3290 * Mark the entry in case the map lock is released. (See
3293 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3294 entry->wiring_thread == NULL,
3295 ("owned map entry %p", entry));
3296 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3297 entry->wiring_thread = curthread;
3298 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
3299 || (entry->protection & prot) != prot) {
3300 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
3303 rv = KERN_INVALID_ADDRESS;
3306 } else if (entry->wired_count == 0) {
3307 entry->wired_count++;
3309 npages = atop(entry->end - entry->start);
3310 if (user_wire && !vm_map_wire_user_count_add(npages)) {
3311 vm_map_wire_entry_failure(map, entry,
3314 rv = KERN_RESOURCE_SHORTAGE;
3319 * Release the map lock, relying on the in-transition
3320 * mark. Mark the map busy for fork.
3322 saved_start = entry->start;
3323 saved_end = entry->end;
3324 last_timestamp = map->timestamp;
3328 faddr = saved_start;
3331 * Simulate a fault to get the page and enter
3332 * it into the physical map.
3334 if ((rv = vm_fault(map, faddr,
3335 VM_PROT_NONE, VM_FAULT_WIRE, NULL)) !=
3338 } while ((faddr += PAGE_SIZE) < saved_end);
3341 if (last_timestamp + 1 != map->timestamp) {
3343 * Look again for the entry because the map was
3344 * modified while it was unlocked. The entry
3345 * may have been clipped, but NOT merged or
3348 if (!vm_map_lookup_entry(map, saved_start,
3351 ("vm_map_wire: lookup failed"));
3352 first_entry = (entry == first_entry) ?
3354 for (entry = next_entry; entry->end < saved_end;
3355 entry = vm_map_entry_succ(entry)) {
3357 * In case of failure, handle entries
3358 * that were not fully wired here;
3359 * fully wired entries are handled
3362 if (rv != KERN_SUCCESS &&
3364 vm_map_wire_entry_failure(map,
3368 if (rv != KERN_SUCCESS) {
3369 vm_map_wire_entry_failure(map, entry, faddr);
3371 vm_map_wire_user_count_sub(npages);
3375 } else if (!user_wire ||
3376 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3377 entry->wired_count++;
3380 * Check the map for holes in the specified region.
3381 * If holes_ok was specified, skip this check.
3383 next_entry = vm_map_entry_succ(entry);
3385 entry->end < end && next_entry->start > entry->end) {
3387 rv = KERN_INVALID_ADDRESS;
3393 need_wakeup = false;
3394 if (first_entry == NULL &&
3395 !vm_map_lookup_entry(map, start, &first_entry)) {
3396 KASSERT(holes_ok, ("vm_map_wire: lookup failed"));
3397 prev_entry = first_entry;
3398 entry = vm_map_entry_succ(first_entry);
3400 prev_entry = vm_map_entry_pred(first_entry);
3401 entry = first_entry;
3403 for (; entry->start < end;
3404 prev_entry = entry, entry = vm_map_entry_succ(entry)) {
3406 * If holes_ok was specified, an empty
3407 * space in the unwired region could have been mapped
3408 * while the map lock was dropped for faulting in the
3409 * pages or draining MAP_ENTRY_IN_TRANSITION.
3410 * Moreover, another thread could be simultaneously
3411 * wiring this new mapping entry. Detect these cases
3412 * and skip any entries marked as in transition not by us.
3414 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3415 entry->wiring_thread != curthread) {
3417 ("vm_map_wire: !HOLESOK and new/changed entry"));
3421 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) {
3423 } else if (rv == KERN_SUCCESS) {
3425 entry->eflags |= MAP_ENTRY_USER_WIRED;
3426 } else if (entry->wired_count == -1) {
3428 * Wiring failed on this entry. Thus, unwiring is
3431 entry->wired_count = 0;
3432 } else if (!user_wire ||
3433 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3435 * Undo the wiring. Wiring succeeded on this entry
3436 * but failed on a later entry.
3438 if (entry->wired_count == 1) {
3439 vm_map_entry_unwire(map, entry);
3441 vm_map_wire_user_count_sub(
3442 atop(entry->end - entry->start));
3444 entry->wired_count--;
3446 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3447 ("vm_map_wire: in-transition flag missing %p", entry));
3448 KASSERT(entry->wiring_thread == curthread,
3449 ("vm_map_wire: alien wire %p", entry));
3450 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
3451 MAP_ENTRY_WIRE_SKIPPED);
3452 entry->wiring_thread = NULL;
3453 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3454 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3457 vm_map_try_merge_entries(map, prev_entry, entry);
3459 vm_map_try_merge_entries(map, prev_entry, entry);
3468 * Push any dirty cached pages in the address range to their pager.
3469 * If syncio is TRUE, dirty pages are written synchronously.
3470 * If invalidate is TRUE, any cached pages are freed as well.
3472 * If the size of the region from start to end is zero, we are
3473 * supposed to flush all modified pages within the region containing
3474 * start. Unfortunately, a region can be split or coalesced with
3475 * neighboring regions, making it difficult to determine what the
3476 * original region was. Therefore, we approximate this requirement by
3477 * flushing the current region containing start.
3479 * Returns an error if any part of the specified range is not mapped.
3487 boolean_t invalidate)
3489 vm_map_entry_t entry, first_entry, next_entry;
3492 vm_ooffset_t offset;
3493 unsigned int last_timestamp;
3496 vm_map_lock_read(map);
3497 VM_MAP_RANGE_CHECK(map, start, end);
3498 if (!vm_map_lookup_entry(map, start, &first_entry)) {
3499 vm_map_unlock_read(map);
3500 return (KERN_INVALID_ADDRESS);
3501 } else if (start == end) {
3502 start = first_entry->start;
3503 end = first_entry->end;
3506 * Make a first pass to check for user-wired memory and holes.
3508 for (entry = first_entry; entry->start < end; entry = next_entry) {
3510 (entry->eflags & MAP_ENTRY_USER_WIRED) != 0) {
3511 vm_map_unlock_read(map);
3512 return (KERN_INVALID_ARGUMENT);
3514 next_entry = vm_map_entry_succ(entry);
3515 if (end > entry->end &&
3516 entry->end != next_entry->start) {
3517 vm_map_unlock_read(map);
3518 return (KERN_INVALID_ADDRESS);
3523 pmap_remove(map->pmap, start, end);
3527 * Make a second pass, cleaning/uncaching pages from the indicated
3530 for (entry = first_entry; entry->start < end;) {
3531 offset = entry->offset + (start - entry->start);
3532 size = (end <= entry->end ? end : entry->end) - start;
3533 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
3535 vm_map_entry_t tentry;
3538 smap = entry->object.sub_map;
3539 vm_map_lock_read(smap);
3540 (void) vm_map_lookup_entry(smap, offset, &tentry);
3541 tsize = tentry->end - offset;
3544 object = tentry->object.vm_object;
3545 offset = tentry->offset + (offset - tentry->start);
3546 vm_map_unlock_read(smap);
3548 object = entry->object.vm_object;
3550 vm_object_reference(object);
3551 last_timestamp = map->timestamp;
3552 vm_map_unlock_read(map);
3553 if (!vm_object_sync(object, offset, size, syncio, invalidate))
3556 vm_object_deallocate(object);
3557 vm_map_lock_read(map);
3558 if (last_timestamp == map->timestamp ||
3559 !vm_map_lookup_entry(map, start, &entry))
3560 entry = vm_map_entry_succ(entry);
3563 vm_map_unlock_read(map);
3564 return (failed ? KERN_FAILURE : KERN_SUCCESS);
3568 * vm_map_entry_unwire: [ internal use only ]
3570 * Make the region specified by this entry pageable.
3572 * The map in question should be locked.
3573 * [This is the reason for this routine's existence.]
3576 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
3580 VM_MAP_ASSERT_LOCKED(map);
3581 KASSERT(entry->wired_count > 0,
3582 ("vm_map_entry_unwire: entry %p isn't wired", entry));
3584 size = entry->end - entry->start;
3585 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0)
3586 vm_map_wire_user_count_sub(atop(size));
3587 pmap_unwire(map->pmap, entry->start, entry->end);
3588 vm_object_unwire(entry->object.vm_object, entry->offset, size,
3590 entry->wired_count = 0;
3594 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
3597 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
3598 vm_object_deallocate(entry->object.vm_object);
3599 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
3603 * vm_map_entry_delete: [ internal use only ]
3605 * Deallocate the given entry from the target map.
3608 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3611 vm_pindex_t offidxstart, offidxend, size1;
3614 vm_map_entry_unlink(map, entry, UNLINK_MERGE_NONE);
3615 object = entry->object.vm_object;
3617 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3618 MPASS(entry->cred == NULL);
3619 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3620 MPASS(object == NULL);
3621 vm_map_entry_deallocate(entry, map->system_map);
3625 size = entry->end - entry->start;
3628 if (entry->cred != NULL) {
3629 swap_release_by_cred(size, entry->cred);
3630 crfree(entry->cred);
3633 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 || object == NULL) {
3634 entry->object.vm_object = NULL;
3635 } else if ((object->flags & OBJ_ANON) != 0 ||
3636 object == kernel_object) {
3637 KASSERT(entry->cred == NULL || object->cred == NULL ||
3638 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3639 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3640 offidxstart = OFF_TO_IDX(entry->offset);
3641 offidxend = offidxstart + atop(size);
3642 VM_OBJECT_WLOCK(object);
3643 if (object->ref_count != 1 &&
3644 ((object->flags & OBJ_ONEMAPPING) != 0 ||
3645 object == kernel_object)) {
3646 vm_object_collapse(object);
3649 * The option OBJPR_NOTMAPPED can be passed here
3650 * because vm_map_delete() already performed
3651 * pmap_remove() on the only mapping to this range
3654 vm_object_page_remove(object, offidxstart, offidxend,
3656 if (offidxend >= object->size &&
3657 offidxstart < object->size) {
3658 size1 = object->size;
3659 object->size = offidxstart;
3660 if (object->cred != NULL) {
3661 size1 -= object->size;
3662 KASSERT(object->charge >= ptoa(size1),
3663 ("object %p charge < 0", object));
3664 swap_release_by_cred(ptoa(size1),
3666 object->charge -= ptoa(size1);
3670 VM_OBJECT_WUNLOCK(object);
3672 if (map->system_map)
3673 vm_map_entry_deallocate(entry, TRUE);
3675 entry->defer_next = curthread->td_map_def_user;
3676 curthread->td_map_def_user = entry;
3681 * vm_map_delete: [ internal use only ]
3683 * Deallocates the given address range from the target
3687 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3689 vm_map_entry_t entry, next_entry;
3691 VM_MAP_ASSERT_LOCKED(map);
3694 return (KERN_SUCCESS);
3697 * Find the start of the region, and clip it.
3698 * Step through all entries in this region.
3700 for (entry = vm_map_lookup_clip_start(map, start, &entry);
3701 entry->start < end; entry = next_entry) {
3703 * Wait for wiring or unwiring of an entry to complete.
3704 * Also wait for any system wirings to disappear on
3707 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3708 (vm_map_pmap(map) != kernel_pmap &&
3709 vm_map_entry_system_wired_count(entry) != 0)) {
3710 unsigned int last_timestamp;
3711 vm_offset_t saved_start;
3713 saved_start = entry->start;
3714 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3715 last_timestamp = map->timestamp;
3716 (void) vm_map_unlock_and_wait(map, 0);
3718 if (last_timestamp + 1 != map->timestamp) {
3720 * Look again for the entry because the map was
3721 * modified while it was unlocked.
3722 * Specifically, the entry may have been
3723 * clipped, merged, or deleted.
3725 next_entry = vm_map_lookup_clip_start(map,
3726 saved_start, &next_entry);
3731 vm_map_clip_end(map, entry, end);
3732 next_entry = vm_map_entry_succ(entry);
3735 * Unwire before removing addresses from the pmap; otherwise,
3736 * unwiring will put the entries back in the pmap.
3738 if (entry->wired_count != 0)
3739 vm_map_entry_unwire(map, entry);
3742 * Remove mappings for the pages, but only if the
3743 * mappings could exist. For instance, it does not
3744 * make sense to call pmap_remove() for guard entries.
3746 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3747 entry->object.vm_object != NULL)
3748 pmap_remove(map->pmap, entry->start, entry->end);
3750 if (entry->end == map->anon_loc)
3751 map->anon_loc = entry->start;
3754 * Delete the entry only after removing all pmap
3755 * entries pointing to its pages. (Otherwise, its
3756 * page frames may be reallocated, and any modify bits
3757 * will be set in the wrong object!)
3759 vm_map_entry_delete(map, entry);
3761 return (KERN_SUCCESS);
3767 * Remove the given address range from the target map.
3768 * This is the exported form of vm_map_delete.
3771 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3776 VM_MAP_RANGE_CHECK(map, start, end);
3777 result = vm_map_delete(map, start, end);
3783 * vm_map_check_protection:
3785 * Assert that the target map allows the specified privilege on the
3786 * entire address region given. The entire region must be allocated.
3788 * WARNING! This code does not and should not check whether the
3789 * contents of the region is accessible. For example a smaller file
3790 * might be mapped into a larger address space.
3792 * NOTE! This code is also called by munmap().
3794 * The map must be locked. A read lock is sufficient.
3797 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3798 vm_prot_t protection)
3800 vm_map_entry_t entry;
3801 vm_map_entry_t tmp_entry;
3803 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3807 while (start < end) {
3811 if (start < entry->start)
3814 * Check protection associated with entry.
3816 if ((entry->protection & protection) != protection)
3818 /* go to next entry */
3820 entry = vm_map_entry_succ(entry);
3827 * vm_map_copy_swap_object:
3829 * Copies a swap-backed object from an existing map entry to a
3830 * new one. Carries forward the swap charge. May change the
3831 * src object on return.
3834 vm_map_copy_swap_object(vm_map_entry_t src_entry, vm_map_entry_t dst_entry,
3835 vm_offset_t size, vm_ooffset_t *fork_charge)
3837 vm_object_t src_object;
3841 src_object = src_entry->object.vm_object;
3842 charged = ENTRY_CHARGED(src_entry);
3843 if ((src_object->flags & OBJ_ANON) != 0) {
3844 VM_OBJECT_WLOCK(src_object);
3845 vm_object_collapse(src_object);
3846 if ((src_object->flags & OBJ_ONEMAPPING) != 0) {
3847 vm_object_split(src_entry);
3848 src_object = src_entry->object.vm_object;
3850 vm_object_reference_locked(src_object);
3851 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3852 VM_OBJECT_WUNLOCK(src_object);
3854 vm_object_reference(src_object);
3855 if (src_entry->cred != NULL &&
3856 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3857 KASSERT(src_object->cred == NULL,
3858 ("OVERCOMMIT: vm_map_copy_anon_entry: cred %p",
3860 src_object->cred = src_entry->cred;
3861 src_object->charge = size;
3863 dst_entry->object.vm_object = src_object;
3865 cred = curthread->td_ucred;
3867 dst_entry->cred = cred;
3868 *fork_charge += size;
3869 if (!(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3871 src_entry->cred = cred;
3872 *fork_charge += size;
3878 * vm_map_copy_entry:
3880 * Copies the contents of the source entry to the destination
3881 * entry. The entries *must* be aligned properly.
3887 vm_map_entry_t src_entry,
3888 vm_map_entry_t dst_entry,
3889 vm_ooffset_t *fork_charge)
3891 vm_object_t src_object;
3892 vm_map_entry_t fake_entry;
3895 VM_MAP_ASSERT_LOCKED(dst_map);
3897 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3900 if (src_entry->wired_count == 0 ||
3901 (src_entry->protection & VM_PROT_WRITE) == 0) {
3903 * If the source entry is marked needs_copy, it is already
3906 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3907 (src_entry->protection & VM_PROT_WRITE) != 0) {
3908 pmap_protect(src_map->pmap,
3911 src_entry->protection & ~VM_PROT_WRITE);
3915 * Make a copy of the object.
3917 size = src_entry->end - src_entry->start;
3918 if ((src_object = src_entry->object.vm_object) != NULL) {
3919 if (src_object->type == OBJT_DEFAULT ||
3920 src_object->type == OBJT_SWAP) {
3921 vm_map_copy_swap_object(src_entry, dst_entry,
3923 /* May have split/collapsed, reload obj. */
3924 src_object = src_entry->object.vm_object;
3926 vm_object_reference(src_object);
3927 dst_entry->object.vm_object = src_object;
3929 src_entry->eflags |= MAP_ENTRY_COW |
3930 MAP_ENTRY_NEEDS_COPY;
3931 dst_entry->eflags |= MAP_ENTRY_COW |
3932 MAP_ENTRY_NEEDS_COPY;
3933 dst_entry->offset = src_entry->offset;
3934 if (src_entry->eflags & MAP_ENTRY_WRITECNT) {
3936 * MAP_ENTRY_WRITECNT cannot
3937 * indicate write reference from
3938 * src_entry, since the entry is
3939 * marked as needs copy. Allocate a
3940 * fake entry that is used to
3941 * decrement object->un_pager writecount
3942 * at the appropriate time. Attach
3943 * fake_entry to the deferred list.
3945 fake_entry = vm_map_entry_create(dst_map);
3946 fake_entry->eflags = MAP_ENTRY_WRITECNT;
3947 src_entry->eflags &= ~MAP_ENTRY_WRITECNT;
3948 vm_object_reference(src_object);
3949 fake_entry->object.vm_object = src_object;
3950 fake_entry->start = src_entry->start;
3951 fake_entry->end = src_entry->end;
3952 fake_entry->defer_next =
3953 curthread->td_map_def_user;
3954 curthread->td_map_def_user = fake_entry;
3957 pmap_copy(dst_map->pmap, src_map->pmap,
3958 dst_entry->start, dst_entry->end - dst_entry->start,
3961 dst_entry->object.vm_object = NULL;
3962 dst_entry->offset = 0;
3963 if (src_entry->cred != NULL) {
3964 dst_entry->cred = curthread->td_ucred;
3965 crhold(dst_entry->cred);
3966 *fork_charge += size;
3971 * We don't want to make writeable wired pages copy-on-write.
3972 * Immediately copy these pages into the new map by simulating
3973 * page faults. The new pages are pageable.
3975 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3981 * vmspace_map_entry_forked:
3982 * Update the newly-forked vmspace each time a map entry is inherited
3983 * or copied. The values for vm_dsize and vm_tsize are approximate
3984 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3987 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3988 vm_map_entry_t entry)
3990 vm_size_t entrysize;
3993 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3995 entrysize = entry->end - entry->start;
3996 vm2->vm_map.size += entrysize;
3997 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3998 vm2->vm_ssize += btoc(entrysize);
3999 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
4000 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
4001 newend = MIN(entry->end,
4002 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
4003 vm2->vm_dsize += btoc(newend - entry->start);
4004 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
4005 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
4006 newend = MIN(entry->end,
4007 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
4008 vm2->vm_tsize += btoc(newend - entry->start);
4014 * Create a new process vmspace structure and vm_map
4015 * based on those of an existing process. The new map
4016 * is based on the old map, according to the inheritance
4017 * values on the regions in that map.
4019 * XXX It might be worth coalescing the entries added to the new vmspace.
4021 * The source map must not be locked.
4024 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
4026 struct vmspace *vm2;
4027 vm_map_t new_map, old_map;
4028 vm_map_entry_t new_entry, old_entry;
4033 old_map = &vm1->vm_map;
4034 /* Copy immutable fields of vm1 to vm2. */
4035 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
4040 vm2->vm_taddr = vm1->vm_taddr;
4041 vm2->vm_daddr = vm1->vm_daddr;
4042 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
4043 vm_map_lock(old_map);
4045 vm_map_wait_busy(old_map);
4046 new_map = &vm2->vm_map;
4047 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
4048 KASSERT(locked, ("vmspace_fork: lock failed"));
4050 error = pmap_vmspace_copy(new_map->pmap, old_map->pmap);
4052 sx_xunlock(&old_map->lock);
4053 sx_xunlock(&new_map->lock);
4054 vm_map_process_deferred();
4059 new_map->anon_loc = old_map->anon_loc;
4060 new_map->flags |= old_map->flags & (MAP_ASLR | MAP_ASLR_IGNSTART);
4062 VM_MAP_ENTRY_FOREACH(old_entry, old_map) {
4063 if ((old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
4064 panic("vm_map_fork: encountered a submap");
4066 inh = old_entry->inheritance;
4067 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4068 inh != VM_INHERIT_NONE)
4069 inh = VM_INHERIT_COPY;
4072 case VM_INHERIT_NONE:
4075 case VM_INHERIT_SHARE:
4077 * Clone the entry, creating the shared object if
4080 object = old_entry->object.vm_object;
4081 if (object == NULL) {
4082 vm_map_entry_back(old_entry);
4083 object = old_entry->object.vm_object;
4087 * Add the reference before calling vm_object_shadow
4088 * to insure that a shadow object is created.
4090 vm_object_reference(object);
4091 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4092 vm_object_shadow(&old_entry->object.vm_object,
4094 old_entry->end - old_entry->start,
4096 /* Transfer the second reference too. */
4098 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4099 old_entry->cred = NULL;
4102 * As in vm_map_merged_neighbor_dispose(),
4103 * the vnode lock will not be acquired in
4104 * this call to vm_object_deallocate().
4106 vm_object_deallocate(object);
4107 object = old_entry->object.vm_object;
4109 VM_OBJECT_WLOCK(object);
4110 vm_object_clear_flag(object, OBJ_ONEMAPPING);
4111 if (old_entry->cred != NULL) {
4112 KASSERT(object->cred == NULL,
4113 ("vmspace_fork both cred"));
4114 object->cred = old_entry->cred;
4115 object->charge = old_entry->end -
4117 old_entry->cred = NULL;
4121 * Assert the correct state of the vnode
4122 * v_writecount while the object is locked, to
4123 * not relock it later for the assertion
4126 if (old_entry->eflags & MAP_ENTRY_WRITECNT &&
4127 object->type == OBJT_VNODE) {
4128 KASSERT(((struct vnode *)object->
4129 handle)->v_writecount > 0,
4130 ("vmspace_fork: v_writecount %p",
4132 KASSERT(object->un_pager.vnp.
4134 ("vmspace_fork: vnp.writecount %p",
4137 VM_OBJECT_WUNLOCK(object);
4141 * Clone the entry, referencing the shared object.
4143 new_entry = vm_map_entry_create(new_map);
4144 *new_entry = *old_entry;
4145 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
4146 MAP_ENTRY_IN_TRANSITION);
4147 new_entry->wiring_thread = NULL;
4148 new_entry->wired_count = 0;
4149 if (new_entry->eflags & MAP_ENTRY_WRITECNT) {
4150 vm_pager_update_writecount(object,
4151 new_entry->start, new_entry->end);
4153 vm_map_entry_set_vnode_text(new_entry, true);
4156 * Insert the entry into the new map -- we know we're
4157 * inserting at the end of the new map.
4159 vm_map_entry_link(new_map, new_entry);
4160 vmspace_map_entry_forked(vm1, vm2, new_entry);
4163 * Update the physical map
4165 pmap_copy(new_map->pmap, old_map->pmap,
4167 (old_entry->end - old_entry->start),
4171 case VM_INHERIT_COPY:
4173 * Clone the entry and link into the map.
4175 new_entry = vm_map_entry_create(new_map);
4176 *new_entry = *old_entry;
4178 * Copied entry is COW over the old object.
4180 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
4181 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_WRITECNT);
4182 new_entry->wiring_thread = NULL;
4183 new_entry->wired_count = 0;
4184 new_entry->object.vm_object = NULL;
4185 new_entry->cred = NULL;
4186 vm_map_entry_link(new_map, new_entry);
4187 vmspace_map_entry_forked(vm1, vm2, new_entry);
4188 vm_map_copy_entry(old_map, new_map, old_entry,
4189 new_entry, fork_charge);
4190 vm_map_entry_set_vnode_text(new_entry, true);
4193 case VM_INHERIT_ZERO:
4195 * Create a new anonymous mapping entry modelled from
4198 new_entry = vm_map_entry_create(new_map);
4199 memset(new_entry, 0, sizeof(*new_entry));
4201 new_entry->start = old_entry->start;
4202 new_entry->end = old_entry->end;
4203 new_entry->eflags = old_entry->eflags &
4204 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
4205 MAP_ENTRY_WRITECNT | MAP_ENTRY_VN_EXEC);
4206 new_entry->protection = old_entry->protection;
4207 new_entry->max_protection = old_entry->max_protection;
4208 new_entry->inheritance = VM_INHERIT_ZERO;
4210 vm_map_entry_link(new_map, new_entry);
4211 vmspace_map_entry_forked(vm1, vm2, new_entry);
4213 new_entry->cred = curthread->td_ucred;
4214 crhold(new_entry->cred);
4215 *fork_charge += (new_entry->end - new_entry->start);
4221 * Use inlined vm_map_unlock() to postpone handling the deferred
4222 * map entries, which cannot be done until both old_map and
4223 * new_map locks are released.
4225 sx_xunlock(&old_map->lock);
4226 sx_xunlock(&new_map->lock);
4227 vm_map_process_deferred();
4233 * Create a process's stack for exec_new_vmspace(). This function is never
4234 * asked to wire the newly created stack.
4237 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4238 vm_prot_t prot, vm_prot_t max, int cow)
4240 vm_size_t growsize, init_ssize;
4244 MPASS((map->flags & MAP_WIREFUTURE) == 0);
4245 growsize = sgrowsiz;
4246 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
4248 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4249 /* If we would blow our VMEM resource limit, no go */
4250 if (map->size + init_ssize > vmemlim) {
4254 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
4261 static int stack_guard_page = 1;
4262 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
4263 &stack_guard_page, 0,
4264 "Specifies the number of guard pages for a stack that grows");
4267 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4268 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
4270 vm_map_entry_t new_entry, prev_entry;
4271 vm_offset_t bot, gap_bot, gap_top, top;
4272 vm_size_t init_ssize, sgp;
4276 * The stack orientation is piggybacked with the cow argument.
4277 * Extract it into orient and mask the cow argument so that we
4278 * don't pass it around further.
4280 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
4281 KASSERT(orient != 0, ("No stack grow direction"));
4282 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
4285 if (max_ssize == 0 ||
4286 !vm_map_range_valid(map, addrbos, addrbos + max_ssize))
4287 return (KERN_INVALID_ADDRESS);
4288 sgp = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ||
4289 (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 :
4290 (vm_size_t)stack_guard_page * PAGE_SIZE;
4291 if (sgp >= max_ssize)
4292 return (KERN_INVALID_ARGUMENT);
4294 init_ssize = growsize;
4295 if (max_ssize < init_ssize + sgp)
4296 init_ssize = max_ssize - sgp;
4298 /* If addr is already mapped, no go */
4299 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
4300 return (KERN_NO_SPACE);
4303 * If we can't accommodate max_ssize in the current mapping, no go.
4305 if (vm_map_entry_succ(prev_entry)->start < addrbos + max_ssize)
4306 return (KERN_NO_SPACE);
4309 * We initially map a stack of only init_ssize. We will grow as
4310 * needed later. Depending on the orientation of the stack (i.e.
4311 * the grow direction) we either map at the top of the range, the
4312 * bottom of the range or in the middle.
4314 * Note: we would normally expect prot and max to be VM_PROT_ALL,
4315 * and cow to be 0. Possibly we should eliminate these as input
4316 * parameters, and just pass these values here in the insert call.
4318 if (orient == MAP_STACK_GROWS_DOWN) {
4319 bot = addrbos + max_ssize - init_ssize;
4320 top = bot + init_ssize;
4323 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
4325 top = bot + init_ssize;
4327 gap_top = addrbos + max_ssize;
4329 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
4330 if (rv != KERN_SUCCESS)
4332 new_entry = vm_map_entry_succ(prev_entry);
4333 KASSERT(new_entry->end == top || new_entry->start == bot,
4334 ("Bad entry start/end for new stack entry"));
4335 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
4336 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
4337 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
4338 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
4339 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
4340 ("new entry lacks MAP_ENTRY_GROWS_UP"));
4341 if (gap_bot == gap_top)
4342 return (KERN_SUCCESS);
4343 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
4344 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
4345 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
4346 if (rv == KERN_SUCCESS) {
4348 * Gap can never successfully handle a fault, so
4349 * read-ahead logic is never used for it. Re-use
4350 * next_read of the gap entry to store
4351 * stack_guard_page for vm_map_growstack().
4353 if (orient == MAP_STACK_GROWS_DOWN)
4354 vm_map_entry_pred(new_entry)->next_read = sgp;
4356 vm_map_entry_succ(new_entry)->next_read = sgp;
4358 (void)vm_map_delete(map, bot, top);
4364 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
4365 * successfully grow the stack.
4368 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
4370 vm_map_entry_t stack_entry;
4374 vm_offset_t gap_end, gap_start, grow_start;
4375 vm_size_t grow_amount, guard, max_grow;
4376 rlim_t lmemlim, stacklim, vmemlim;
4378 bool gap_deleted, grow_down, is_procstack;
4390 * Disallow stack growth when the access is performed by a
4391 * debugger or AIO daemon. The reason is that the wrong
4392 * resource limits are applied.
4394 if (p != initproc && (map != &p->p_vmspace->vm_map ||
4395 p->p_textvp == NULL))
4396 return (KERN_FAILURE);
4398 MPASS(!map->system_map);
4400 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
4401 stacklim = lim_cur(curthread, RLIMIT_STACK);
4402 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4404 /* If addr is not in a hole for a stack grow area, no need to grow. */
4405 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
4406 return (KERN_FAILURE);
4407 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
4408 return (KERN_SUCCESS);
4409 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
4410 stack_entry = vm_map_entry_succ(gap_entry);
4411 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
4412 stack_entry->start != gap_entry->end)
4413 return (KERN_FAILURE);
4414 grow_amount = round_page(stack_entry->start - addr);
4416 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
4417 stack_entry = vm_map_entry_pred(gap_entry);
4418 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
4419 stack_entry->end != gap_entry->start)
4420 return (KERN_FAILURE);
4421 grow_amount = round_page(addr + 1 - stack_entry->end);
4424 return (KERN_FAILURE);
4426 guard = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ||
4427 (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 :
4428 gap_entry->next_read;
4429 max_grow = gap_entry->end - gap_entry->start;
4430 if (guard > max_grow)
4431 return (KERN_NO_SPACE);
4433 if (grow_amount > max_grow)
4434 return (KERN_NO_SPACE);
4437 * If this is the main process stack, see if we're over the stack
4440 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
4441 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
4442 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
4443 return (KERN_NO_SPACE);
4448 if (is_procstack && racct_set(p, RACCT_STACK,
4449 ctob(vm->vm_ssize) + grow_amount)) {
4451 return (KERN_NO_SPACE);
4457 grow_amount = roundup(grow_amount, sgrowsiz);
4458 if (grow_amount > max_grow)
4459 grow_amount = max_grow;
4460 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
4461 grow_amount = trunc_page((vm_size_t)stacklim) -
4467 limit = racct_get_available(p, RACCT_STACK);
4469 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
4470 grow_amount = limit - ctob(vm->vm_ssize);
4473 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
4474 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
4481 if (racct_set(p, RACCT_MEMLOCK,
4482 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
4492 /* If we would blow our VMEM resource limit, no go */
4493 if (map->size + grow_amount > vmemlim) {
4500 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
4509 if (vm_map_lock_upgrade(map)) {
4511 vm_map_lock_read(map);
4516 grow_start = gap_entry->end - grow_amount;
4517 if (gap_entry->start + grow_amount == gap_entry->end) {
4518 gap_start = gap_entry->start;
4519 gap_end = gap_entry->end;
4520 vm_map_entry_delete(map, gap_entry);
4523 MPASS(gap_entry->start < gap_entry->end - grow_amount);
4524 vm_map_entry_resize(map, gap_entry, -grow_amount);
4525 gap_deleted = false;
4527 rv = vm_map_insert(map, NULL, 0, grow_start,
4528 grow_start + grow_amount,
4529 stack_entry->protection, stack_entry->max_protection,
4530 MAP_STACK_GROWS_DOWN);
4531 if (rv != KERN_SUCCESS) {
4533 rv1 = vm_map_insert(map, NULL, 0, gap_start,
4534 gap_end, VM_PROT_NONE, VM_PROT_NONE,
4535 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
4536 MPASS(rv1 == KERN_SUCCESS);
4538 vm_map_entry_resize(map, gap_entry,
4542 grow_start = stack_entry->end;
4543 cred = stack_entry->cred;
4544 if (cred == NULL && stack_entry->object.vm_object != NULL)
4545 cred = stack_entry->object.vm_object->cred;
4546 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
4548 /* Grow the underlying object if applicable. */
4549 else if (stack_entry->object.vm_object == NULL ||
4550 vm_object_coalesce(stack_entry->object.vm_object,
4551 stack_entry->offset,
4552 (vm_size_t)(stack_entry->end - stack_entry->start),
4553 grow_amount, cred != NULL)) {
4554 if (gap_entry->start + grow_amount == gap_entry->end) {
4555 vm_map_entry_delete(map, gap_entry);
4556 vm_map_entry_resize(map, stack_entry,
4559 gap_entry->start += grow_amount;
4560 stack_entry->end += grow_amount;
4562 map->size += grow_amount;
4567 if (rv == KERN_SUCCESS && is_procstack)
4568 vm->vm_ssize += btoc(grow_amount);
4571 * Heed the MAP_WIREFUTURE flag if it was set for this process.
4573 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
4574 rv = vm_map_wire_locked(map, grow_start,
4575 grow_start + grow_amount,
4576 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
4578 vm_map_lock_downgrade(map);
4582 if (racct_enable && rv != KERN_SUCCESS) {
4584 error = racct_set(p, RACCT_VMEM, map->size);
4585 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
4587 error = racct_set(p, RACCT_MEMLOCK,
4588 ptoa(pmap_wired_count(map->pmap)));
4589 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
4591 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
4592 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
4601 * Unshare the specified VM space for exec. If other processes are
4602 * mapped to it, then create a new one. The new vmspace is null.
4605 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
4607 struct vmspace *oldvmspace = p->p_vmspace;
4608 struct vmspace *newvmspace;
4610 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
4611 ("vmspace_exec recursed"));
4612 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit);
4613 if (newvmspace == NULL)
4615 newvmspace->vm_swrss = oldvmspace->vm_swrss;
4617 * This code is written like this for prototype purposes. The
4618 * goal is to avoid running down the vmspace here, but let the
4619 * other process's that are still using the vmspace to finally
4620 * run it down. Even though there is little or no chance of blocking
4621 * here, it is a good idea to keep this form for future mods.
4623 PROC_VMSPACE_LOCK(p);
4624 p->p_vmspace = newvmspace;
4625 PROC_VMSPACE_UNLOCK(p);
4626 if (p == curthread->td_proc)
4627 pmap_activate(curthread);
4628 curthread->td_pflags |= TDP_EXECVMSPC;
4633 * Unshare the specified VM space for forcing COW. This
4634 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
4637 vmspace_unshare(struct proc *p)
4639 struct vmspace *oldvmspace = p->p_vmspace;
4640 struct vmspace *newvmspace;
4641 vm_ooffset_t fork_charge;
4643 if (oldvmspace->vm_refcnt == 1)
4646 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4647 if (newvmspace == NULL)
4649 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4650 vmspace_free(newvmspace);
4653 PROC_VMSPACE_LOCK(p);
4654 p->p_vmspace = newvmspace;
4655 PROC_VMSPACE_UNLOCK(p);
4656 if (p == curthread->td_proc)
4657 pmap_activate(curthread);
4658 vmspace_free(oldvmspace);
4665 * Finds the VM object, offset, and
4666 * protection for a given virtual address in the
4667 * specified map, assuming a page fault of the
4670 * Leaves the map in question locked for read; return
4671 * values are guaranteed until a vm_map_lookup_done
4672 * call is performed. Note that the map argument
4673 * is in/out; the returned map must be used in
4674 * the call to vm_map_lookup_done.
4676 * A handle (out_entry) is returned for use in
4677 * vm_map_lookup_done, to make that fast.
4679 * If a lookup is requested with "write protection"
4680 * specified, the map may be changed to perform virtual
4681 * copying operations, although the data referenced will
4685 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4687 vm_prot_t fault_typea,
4688 vm_map_entry_t *out_entry, /* OUT */
4689 vm_object_t *object, /* OUT */
4690 vm_pindex_t *pindex, /* OUT */
4691 vm_prot_t *out_prot, /* OUT */
4692 boolean_t *wired) /* OUT */
4694 vm_map_entry_t entry;
4695 vm_map_t map = *var_map;
4697 vm_prot_t fault_type;
4698 vm_object_t eobject;
4704 vm_map_lock_read(map);
4708 * Lookup the faulting address.
4710 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4711 vm_map_unlock_read(map);
4712 return (KERN_INVALID_ADDRESS);
4720 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4721 vm_map_t old_map = map;
4723 *var_map = map = entry->object.sub_map;
4724 vm_map_unlock_read(old_map);
4729 * Check whether this task is allowed to have this page.
4731 prot = entry->protection;
4732 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4733 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4734 if (prot == VM_PROT_NONE && map != kernel_map &&
4735 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4736 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4737 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4738 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4739 goto RetryLookupLocked;
4741 fault_type = fault_typea & VM_PROT_ALL;
4742 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4743 vm_map_unlock_read(map);
4744 return (KERN_PROTECTION_FAILURE);
4746 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4747 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4748 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4749 ("entry %p flags %x", entry, entry->eflags));
4750 if ((fault_typea & VM_PROT_COPY) != 0 &&
4751 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4752 (entry->eflags & MAP_ENTRY_COW) == 0) {
4753 vm_map_unlock_read(map);
4754 return (KERN_PROTECTION_FAILURE);
4758 * If this page is not pageable, we have to get it for all possible
4761 *wired = (entry->wired_count != 0);
4763 fault_type = entry->protection;
4764 size = entry->end - entry->start;
4767 * If the entry was copy-on-write, we either ...
4769 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4771 * If we want to write the page, we may as well handle that
4772 * now since we've got the map locked.
4774 * If we don't need to write the page, we just demote the
4775 * permissions allowed.
4777 if ((fault_type & VM_PROT_WRITE) != 0 ||
4778 (fault_typea & VM_PROT_COPY) != 0) {
4780 * Make a new object, and place it in the object
4781 * chain. Note that no new references have appeared
4782 * -- one just moved from the map to the new
4785 if (vm_map_lock_upgrade(map))
4788 if (entry->cred == NULL) {
4790 * The debugger owner is charged for
4793 cred = curthread->td_ucred;
4795 if (!swap_reserve_by_cred(size, cred)) {
4798 return (KERN_RESOURCE_SHORTAGE);
4802 eobject = entry->object.vm_object;
4803 vm_object_shadow(&entry->object.vm_object,
4804 &entry->offset, size, entry->cred, false);
4805 if (eobject == entry->object.vm_object) {
4807 * The object was not shadowed.
4809 swap_release_by_cred(size, entry->cred);
4810 crfree(entry->cred);
4813 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4815 vm_map_lock_downgrade(map);
4818 * We're attempting to read a copy-on-write page --
4819 * don't allow writes.
4821 prot &= ~VM_PROT_WRITE;
4826 * Create an object if necessary.
4828 if (entry->object.vm_object == NULL && !map->system_map) {
4829 if (vm_map_lock_upgrade(map))
4831 entry->object.vm_object = vm_object_allocate_anon(atop(size),
4832 NULL, entry->cred, entry->cred != NULL ? size : 0);
4835 vm_map_lock_downgrade(map);
4839 * Return the object/offset from this entry. If the entry was
4840 * copy-on-write or empty, it has been fixed up.
4842 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4843 *object = entry->object.vm_object;
4846 return (KERN_SUCCESS);
4850 * vm_map_lookup_locked:
4852 * Lookup the faulting address. A version of vm_map_lookup that returns
4853 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4856 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4858 vm_prot_t fault_typea,
4859 vm_map_entry_t *out_entry, /* OUT */
4860 vm_object_t *object, /* OUT */
4861 vm_pindex_t *pindex, /* OUT */
4862 vm_prot_t *out_prot, /* OUT */
4863 boolean_t *wired) /* OUT */
4865 vm_map_entry_t entry;
4866 vm_map_t map = *var_map;
4868 vm_prot_t fault_type = fault_typea;
4871 * Lookup the faulting address.
4873 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4874 return (KERN_INVALID_ADDRESS);
4879 * Fail if the entry refers to a submap.
4881 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4882 return (KERN_FAILURE);
4885 * Check whether this task is allowed to have this page.
4887 prot = entry->protection;
4888 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4889 if ((fault_type & prot) != fault_type)
4890 return (KERN_PROTECTION_FAILURE);
4893 * If this page is not pageable, we have to get it for all possible
4896 *wired = (entry->wired_count != 0);
4898 fault_type = entry->protection;
4900 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4902 * Fail if the entry was copy-on-write for a write fault.
4904 if (fault_type & VM_PROT_WRITE)
4905 return (KERN_FAILURE);
4907 * We're attempting to read a copy-on-write page --
4908 * don't allow writes.
4910 prot &= ~VM_PROT_WRITE;
4914 * Fail if an object should be created.
4916 if (entry->object.vm_object == NULL && !map->system_map)
4917 return (KERN_FAILURE);
4920 * Return the object/offset from this entry. If the entry was
4921 * copy-on-write or empty, it has been fixed up.
4923 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4924 *object = entry->object.vm_object;
4927 return (KERN_SUCCESS);
4931 * vm_map_lookup_done:
4933 * Releases locks acquired by a vm_map_lookup
4934 * (according to the handle returned by that lookup).
4937 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4940 * Unlock the main-level map
4942 vm_map_unlock_read(map);
4946 vm_map_max_KBI(const struct vm_map *map)
4949 return (vm_map_max(map));
4953 vm_map_min_KBI(const struct vm_map *map)
4956 return (vm_map_min(map));
4960 vm_map_pmap_KBI(vm_map_t map)
4967 vm_map_range_valid_KBI(vm_map_t map, vm_offset_t start, vm_offset_t end)
4970 return (vm_map_range_valid(map, start, end));
4975 _vm_map_assert_consistent(vm_map_t map, int check)
4977 vm_map_entry_t entry, prev;
4978 vm_map_entry_t cur, header, lbound, ubound;
4979 vm_size_t max_left, max_right;
4984 if (enable_vmmap_check != check)
4987 header = prev = &map->header;
4988 VM_MAP_ENTRY_FOREACH(entry, map) {
4989 KASSERT(prev->end <= entry->start,
4990 ("map %p prev->end = %jx, start = %jx", map,
4991 (uintmax_t)prev->end, (uintmax_t)entry->start));
4992 KASSERT(entry->start < entry->end,
4993 ("map %p start = %jx, end = %jx", map,
4994 (uintmax_t)entry->start, (uintmax_t)entry->end));
4995 KASSERT(entry->left == header ||
4996 entry->left->start < entry->start,
4997 ("map %p left->start = %jx, start = %jx", map,
4998 (uintmax_t)entry->left->start, (uintmax_t)entry->start));
4999 KASSERT(entry->right == header ||
5000 entry->start < entry->right->start,
5001 ("map %p start = %jx, right->start = %jx", map,
5002 (uintmax_t)entry->start, (uintmax_t)entry->right->start));
5004 lbound = ubound = header;
5006 if (entry->start < cur->start) {
5009 KASSERT(cur != lbound,
5010 ("map %p cannot find %jx",
5011 map, (uintmax_t)entry->start));
5012 } else if (cur->end <= entry->start) {
5015 KASSERT(cur != ubound,
5016 ("map %p cannot find %jx",
5017 map, (uintmax_t)entry->start));
5019 KASSERT(cur == entry,
5020 ("map %p cannot find %jx",
5021 map, (uintmax_t)entry->start));
5025 max_left = vm_map_entry_max_free_left(entry, lbound);
5026 max_right = vm_map_entry_max_free_right(entry, ubound);
5027 KASSERT(entry->max_free == vm_size_max(max_left, max_right),
5028 ("map %p max = %jx, max_left = %jx, max_right = %jx", map,
5029 (uintmax_t)entry->max_free,
5030 (uintmax_t)max_left, (uintmax_t)max_right));
5033 KASSERT(prev->end <= entry->start,
5034 ("map %p prev->end = %jx, start = %jx", map,
5035 (uintmax_t)prev->end, (uintmax_t)entry->start));
5039 #include "opt_ddb.h"
5041 #include <sys/kernel.h>
5043 #include <ddb/ddb.h>
5046 vm_map_print(vm_map_t map)
5048 vm_map_entry_t entry, prev;
5050 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
5052 (void *)map->pmap, map->nentries, map->timestamp);
5055 prev = &map->header;
5056 VM_MAP_ENTRY_FOREACH(entry, map) {
5057 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
5058 (void *)entry, (void *)entry->start, (void *)entry->end,
5061 static const char * const inheritance_name[4] =
5062 {"share", "copy", "none", "donate_copy"};
5064 db_iprintf(" prot=%x/%x/%s",
5066 entry->max_protection,
5067 inheritance_name[(int)(unsigned char)
5068 entry->inheritance]);
5069 if (entry->wired_count != 0)
5070 db_printf(", wired");
5072 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
5073 db_printf(", share=%p, offset=0x%jx\n",
5074 (void *)entry->object.sub_map,
5075 (uintmax_t)entry->offset);
5076 if (prev == &map->header ||
5077 prev->object.sub_map !=
5078 entry->object.sub_map) {
5080 vm_map_print((vm_map_t)entry->object.sub_map);
5084 if (entry->cred != NULL)
5085 db_printf(", ruid %d", entry->cred->cr_ruid);
5086 db_printf(", object=%p, offset=0x%jx",
5087 (void *)entry->object.vm_object,
5088 (uintmax_t)entry->offset);
5089 if (entry->object.vm_object && entry->object.vm_object->cred)
5090 db_printf(", obj ruid %d charge %jx",
5091 entry->object.vm_object->cred->cr_ruid,
5092 (uintmax_t)entry->object.vm_object->charge);
5093 if (entry->eflags & MAP_ENTRY_COW)
5094 db_printf(", copy (%s)",
5095 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
5098 if (prev == &map->header ||
5099 prev->object.vm_object !=
5100 entry->object.vm_object) {
5102 vm_object_print((db_expr_t)(intptr_t)
5103 entry->object.vm_object,
5113 DB_SHOW_COMMAND(map, map)
5117 db_printf("usage: show map <addr>\n");
5120 vm_map_print((vm_map_t)addr);
5123 DB_SHOW_COMMAND(procvm, procvm)
5128 p = db_lookup_proc(addr);
5133 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
5134 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
5135 (void *)vmspace_pmap(p->p_vmspace));
5137 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);